This selection of D. H. Mellor’s papers demonstrates the wide-ranging originality of his work. It gathers together sixteen major papers on related topics written over the last seventeen years. Together they form a complete modern metaphysics. The first five papers are on aspects of the mind: on our ‘selves’, their supposed subjectivity and how we refer to them, on the nature of conscious belief and on computational and physicalist theories of the mind. The next five papers deal with dispositions, natural kinds, laws of nature and how they involve natural necessity, universals and objective chances, and the relation between properties and predicates. Then follow three papers about the relations between time, change and causation, the nature of individual causes and effects and of the causal relation between them, and how causation depends on chance. The last three papers discuss the relation between chance and degrees of belief, give a solution to the problem of induction, and argue for an objective interpretation of decision theory. Two of the papers included here have been especially written for this volume, another has been revised for it, and many have hitherto been relatively inaccessible. A substantial introduction summarises the papers and indicates the connections between them.Matters of MetaphysicsMatters of Metaphysics D. H. MELLOR Professor of Philosophy, University of Cambridge CAMBRIDGE UNIVERSITY PRESS Cambridge New York Port Chester Melbourne SydneyCAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo ‘Cambridge University Press ‘The Edinburgh Building, Cambridge CB2 2RU, UK, Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521411172 © Cambridge University Press 1991 This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 1991 A catalogue record for this publication is available from the British Library ISBN-13 978-0-521-41117-2 hardback ISBN-10 0-521-41117-3 hardback Transferred to digital printing 2005For Tim Crane and Jamie WhyteContents Preface Introduction. Minds Analytic philosophy and the self I and now (1989) Consciousness and degrees of belief (1980) How much of the mind is a computer? (1988) (with Tim Crane) There is no question of physicalism (1990) Properties and laws In defence of dispositions (1974) Natural kinds (1977) Necessities and universals in natural laws (1980) Laws, chances and properties (1990) Properties and predicates Causation McTaggart, fixity and coming true (1981) The singularly affecting facts of causation (1987) On raising the chances of effects (1988) Prediction and decision Chance and degrees of belief (1982) The warrant of induction (1988) Objective decision making (1983) References page xi xv ld 30 61 82 104 123 136 154 170 183 201 225 235 254 269 288Preface Two of the papers in this volume, chapters 1 and 10, are published here for the first time. Chapter 1 is also due to appear in German as: “Analytische Philosophie und das Selbst’, Orientierung durch Philosophie, ed. P. Koslowski, 1991, Tiibingen: J. C. B. Mohr (Paul Siebeck). Chapter 10, ‘Properties and predicates’, is also due to appear, with a reply by D. M. Armstrong, in: Ontology, Causality, and Mind, ed. K. Campbell et al., Cambridge: Cambridge University Press. Chapter 9 has been significantly revised from its first publication as: ‘Laws, chances and properties’, International Studies in the Philosophy of Science 4 (1990), 159-70. Reprinted by permission of the Editors. The other chapters are reprinted here with no substantial changes. The details of their original publication are: Chapter 2: ‘I and now’, Proceedings of the Aristotelian Society 89 (1989), 79-94. © The Aristotelian Society 1989. Reprinted by courtesy of the Editor. Chapter 3: ‘Consciousness and degrees of belief’, Prospects for Pragmatism, ed. D. H. Mellor, Cambridge: Cambridge University Press (1980), 139- B. Chapter 4: ‘How much of the mind is a computer?”, Computers, Brains and Minds, ed. P. Slezak and W. R. Albury, Dordrecht: Kluwer (1988), 47- 69. © 1989 by Kluwer Academic Publishers. Reprinted by permission of Kluwer Academic Publishers.xii Preface Chapter 5 (with Tim Crane): ‘There is no question of physicalism’, Mind 99 (1990), 185-206. Reprinted by permission of Tim Crane and Oxford University Press. Chapter 6: ‘In defense of dispositions’, Philosophical Review 83 (1974), 157- 81. Reprinted by permission of the publisher. Chapter 7: ‘Natural kinds’, British Journal for the Philosophy of Science 28 (1977), 299-312. Reprinted by permission of the Editor. Chapter 8: ‘Necessities and universals in natural laws’, Science, Belief and Behaviour, ed. D. H. Mellor, Cambridge: Cambridge University Press (1980), 105-25. Chapter 11: ‘McTaggart, fixity and coming true’, Reduction, Time and Reality, ed. R. Healey, Cambridge: Cambridge University Press (1981), 79-91. Chapter 12: ‘The singularly affecting facts of causation’, Metaphysics and Morality: Essays in Honour of J. J. C. Smart, ed. P. Pettit et al., Oxford: Blackwell (1987), 111-36. Chapter 13: ‘On raising the chances of effects’, Probability and Causality, ed. J. H. Fetzer, Dordrecht: Reidel (1988), 229-39. © 1988 by D. Reidel Publishing Company. Reprinted by permission of Kluwer Academic Publishers. Chapter 14: ‘Chance and degrees of belief’, What? Where? When? Why?, ed. R. McLaughlin, Dordrecht: Reidel (1982), 49-68. Copyright © 1982 by D. Reidel Publishing Company. Reprinted by permission of Kluwer Academic Publishers. Chapter 15: The Warrant of Induction, Cambridge: Cambridge University Press (1988). Chapter 16: ‘Objective decision making’, Social Theory and Practice 9 (1983), 289-309. I am grateful to those concerned for permission to republish this material in this book, and to Cambridge University Press for enabling me to do so. For the material itself | am beholden to many more people than I can mention here. I acknowledge specific debts in references to works listed at the end of the book; but since the bibliographies and acknowledgements of my previouslyPreface xiii published articles have not been reprinted, I must emphasise that I am indebted to far more of the literature than I refer to. But my deepest debts are to those with whom I have discussed these subjects for over twenty-five years. I owe much to many philosophers I have met elsewhere in Britain, Australasia (whose hard thinking and straight talking I find especially congenial), the United States, Ireland, Germany, Sweden, Russia, Yugoslavia and Holland. But I owe most to my colleagues and students in Cambridge, who have taught more than I can say about philosophy and how to do it. Though I cannot name them all, I thank them all; and I hope they will let the name of my mentor, friend and philosophical exemplar, Richard Braithwaite, who died last year, stand for all I owe to Cambridge philosophy. Cambridge March 1991Introduction The papers that follow have been written over a period of twenty years and are those which I now think most worth preserving, both individually and collectively. They contain inevitable repetitions, but also important connections, and between them constitute a metaphysics which I hope will interest other philosophers and students of the subject. As they were written at different times and for different purposes, they naturally vary in depth, detail and style, variations which | have not tried to iron out. In particular, I have kept the style of the two chapters (1 and 15) originally written as lectures. The first of these, ‘Analytic philosophy and the self”, was given in Leibniz’ house in Hannover on 16 November 1989 in a series of public lectures on the réle of philosophy today. It starts with a statement of what I take to be the main uses and virtues of analytic philosophy, and that made it seem a good first chapter for this book. After this, however, readers will find nothing more about philosophy itself, since the philosophy of philosophy seems to me among its least enlightening branches. In the sense in which astronomers are interested not in astronomy but in the stars, I am interested not in philosophy but in the various philosophical topics dealt with in this book — topics on which I find discussions of what philosophy is and how to do it shed very little light. I think the proof of our methods lies rather in the results of applying them, and my case for my method, such as it is, rests on the contents of the ensuing chapters. Minds The first five chapters concern aspects of the mind. Chapters 1 and 2 contain work on the self motivated by scepticism about recent claims for its subjectivity. In chapter 1, I attack the myth of subjectivity by showing how a wholly objective world can incorporate all our apparently subjective first- person knowledge of ourselves and of the world as it appears from our own perspective. Chapter 2, ‘I and now’ (1989), supplements chapter 1 by showing how the causal mechanisms of our minds enable our subjective — first-person present tense — thought and speech to refer directly and infallibly to ourselves and to the present time. This remarkable ability turns out to need neither subjective facts or selves, internal representations of ourselves or of the xvxvi Introduction present, nor any problematic capacity for (or concepts of) infallible self- reference or self-knowledge. All it needs is causal contiguity: the fact that causes are contiguous to their immediate effects. That is all the mental machinery it takes to put us and the present time into the truth conditions of our subjective statements and beliefs. Subjective beliefs figure also in the theory of conscious belief developed in chapter 3, ‘Consciousness and degrees of belief’ (1980), written for a volume of essays to commemorate the fiftieth anniversary of the death of F. P. Ramsey. Our direct if fallible awareness of what we believe poses a problem for the functionalist accounts of beliefs pioneered by Ramsey, i.e. accounts which define beliefs by how our senses cause them, how they interact with each other and with desires and other states of mind, and how they combine with desires to cause actions. The problem is to say what conscious belief (which I call assent) is, how it arises, and how it relates to its unconscious counterpart. I argue that assenting is believing one believes, a ‘second order’ subjective belief caused by one’s ‘first order’ beliefs via an inner sense. I show how this accommodates degrees of both belief and assent — and justifies taking subjective probability to measure actual, rather than merely rational, degrees of belief. Finally I tackle the apparent and obviously false consequence (C) that assenting to any proposition entails assenting to every proposition which it entails and of which we're conscious. My treatment of this problem here is inconclusive, but I now think I could accommodate (C)’s falsity by taking account of limitations in the causal mechanisms of belief formation discussed in chapter 4. But that account remains to be given, as does an answer to recent apparent counter-examples to my equation of assent with second order belief — an equation which, however, failing any other account of assent, I still think I believe. Chapter 4, ‘How much of the mind is a computer?’ (1988), was written in response to some extravagant claims for computational psychology. In it I first consider what it might usefully mean to claim that mental processes are computations, i.e. causal transformations of information embodied in representations. | argue that this claim is vacuous unless computation is understood as part of a process of forming beliefs, but that this does not preclude non-trivial information-processing theories of perception and inference. However, I then argue that there can be no such theories of mental processes involving other mental states such as desire which, unlike beliefs, are not essentially truth-seeking. Most of the mind, I therefore conclude, is not a computer.Introduction xvii In chapter 5, ‘There is no question of physicalism’ (1990), Tim Crane and I attack the common but unclear assumption in much recent philosophy of mind that mental states and processes are all really physical. To avoid vacuity, this thesis needs a non-question-begging definition of ‘physical’ which will prevent the mental counting as physical in its own right; and we argue that no such definition exists. First we show why the physicalist cannot define the physical either as what reduces to physics, or as what has causes or effects. Then we show why he cannot define it as the non-intentional or as the law-governed: since all the problematic features of intentionality occur also in physics, and laws occur also in psychology. Finally we argue that the mental does not even supervene on the physical, and conclude that the whole question of physicalism is as trivial as the doctrine itself is false. Properties and laws Chapters 6 to 10 concern natural properties, kinds and laws. Getting the right account of these is important not only in itself, but for the light it sheds on other metaphysical matters, including the metaphysics of the mind. Thus the misconception that laws are necessary has led many to deny the psychological and psychophysical laws discussed in chapter 5. Conversely, many attempts to reduce mental states to behavioural dispositions have been motivated by the anti-realist view of dispositions which I attack in chapter 6. In this chapter, ‘In defence of dispositions’ (1974), I argue that dispositions, both physical and mental, are real properties of things and people: i.e. that changes in them have real causes and effects. I argue moreover that dispositions have these causal powers in their own right, needing no non- dispositional properties to give them a real basis, such as molecular structures are supposed to provide for solubility, or brain states for mental dispositions like beliefs and desires. For these properties too are only dispositions: like inertial masses, which are no less real for being nothing but dispositions which objects have to accelerate under applied forces. In chapter 7, ‘Natural kinds’ (1977), I turn from the supposed bases of dispositions to such supposedly essential properties of natural kinds as being H,0, which essentialists say that water not only is but must be. The paper was written to rebut two well-known semantic arguments for these essences, one based on an anti-Fregean account of the extension of terms like ‘water’, the other on the necessary truth of identity statements like ‘water is H,O’. Both arguments fail: the first by having premises which both fail to be true of real kind terms and fail to entail essentialism; the second by begging the question, since ‘water is H,O’ won’t even be true, never mind necessary, unless being H,0 is an essential property of water — which, I argue, it isn’t. And as forxviii Introduction H,0, so in general: apparently essential properties are either not even shared by every actual thing of the kind, or their importance as properties is evidently more a feature of our theories than of the world. Chapter 8, ‘Necessities and universals in natural laws’ (1980), was written for a festschrift for Richard Braithwaite, complementing a paper supporting Braithwaite’s Humean view of laws by attacking two non-Humean alternatives. One of these takes laws to be metaphysically necessary, the other takes them to be second order relations between the universals (properties and relations) involved. The implausible metaphysics of these views could be justified, I maintain, only by yielding better solutions than Humeans can provide to such problems as how an uninstantiated law (like Newton's first law of motion) can hold non-trivially - which, I argue, they fail to do. In chapter 9, ‘Laws, chances and properties’ (1990, revised), I supplement the previous chapter’s critique of rival views with a unified account of both deterministic and indeterministic laws, based on the probabilities which, I argue, all laws contain. As in my (1971) The Matter of Chance, | take these probabilities to be real single-case chances, an interpretation modified and further defended in chapter 14. I then identify the ‘natural necessity’ of deterministic laws with the chance of 1 which they give their corresponding regularities. Laws in general I take to be embodied in the actual properties and relations (including chances) they contain, where actual properties and relations are those that would be quantified over by the Ramsey sentence of the conjunction of all true law statements. Chapter 10, ‘Properties and predicates’, looks at how closely universals characterised non-semantically — as in chapter 9 — correspond to our general concepts, i.e. to the meanings of our predicates. I argue that they correspond hardly at all. I show how even if the effects on our senses of a single property fixed the extension of a predicate like ‘red’, the property would still not fix (though its laws would constrain) the predicate’s connotations. I argue moreover that in fact few if any predicates, except some of those used to state natural laws, even correspond to single universals, let alone have them as anything like their meanings. Causation In chapters 11 to 13 I turn from properties and laws to causation, and first to the link between causation and time. In chapter 11, ‘McTaggart, fixity and coming true’ (1981), I first give a tenseless account of change, using arguments which are taken further in my (1981) Real Time. | then use this to attack the idea that causation gets its temporal direction from the flow of time, which makes events necessary when they or their sufficient causes becomeIntroduction — xix ‘fixed’ (or the tenseless propositions saying they or their causes occur ‘come true’) by becoming present. I argue that these concepts of ‘fixity’ and ‘coming true’ either entail the contradictions that McTaggart showed to be inherent in the flow of time, or reduce to tenseless temporal relations which fail to give causation its temporal direction. Chapter 12, ‘The singularly affecting facts of causation’ (1987), is about the relata of causation. In it I argue that singular causation primarily links facts rather than particular events. I rebut a well-known argument to the contrary, and show how causation linking events (Don’s fall caused his death) follows from causation linking existential facts about them (Don died because he fel , there was a death of Don because there was a fall of Don). I argue moreover that the temporal order which (I argue in Real Time) causal order entails therefore also primarily links facts rather than events or spacetime regions; and infer from this that spacetime regions are not events but are individuated by the spatio-temporal relations of facts located at them. In chapter 13, ‘On raising the chances of effects’ (1988), I argue that although causes need not determine their effects, they must at least raise their chances. The basis of the argument is what I call causation’s ‘means-end’ connotation: namely, that when effects are ends, their causes are ipso facto means to them. I define the means-end relation without appeal to causation by using the expected utility principle of non-causal decision theory — requiring only that the probabilities involved be objective chances rather than subjective probabilities - and hence derive the required result. I then show how to read ‘raise the chances’ non-causally, as saying that effects’ chances are greater in the circumstances with their causes than without them, dealing with apparent counter-examples by showing how these conditionals can be made true by probabilistic dispositions. Finally I show how a cause’s efficacy increases the more it raises its effects’ chances, thus explaining our prejudice in favour of deterministic causation. Prediction and decision In the last three papers I defend my concept of objective chance and show how it affects what we should believe, predict and decide to do. Chapter 14, ‘Chance and degrees of belief’ (1982), was written in response to a critique by Wesley Salmon of my belief-based version of the so-called propensity theory of chance. In it I develop the analogy between chance and secondary qualities like colour, both of which are dispositional properties of events or objects defined by the mental states (degrees of belief, visual sensations) that constitute veridical perceptions of them. However, drawing on the arguments of chapter 6, I deny that this shows chances not toxx Introduction be real properties, but argue that it does show how their existence, and the rationality of acting on them, can be compatible with an underlying determinism. Chapter 15, ‘The warrant of induction’ (1988), was my Inaugural Lecture as Professor of Philosophy at Cambridge University. This seemed an appropriate occasion to work out and defend the inductive solution to the problem of induction suggested by Ramsey and further developed by Braithwaite and others. In it I defend inductive solutions against charges of vicious circularity by arguing generally that what warrants a belief is its having a very high chance of being true, whether or not the believer knows it has. I then apply this criterion to beliefs formed by inference, and show how the habit of forming inferential habits by induction warrants them by making them most likely to be formed when the conclusion of the inference has a high chance of being true if the premise is true. Chapter 16, ‘Objective decision making’ (1983), defends an objective reading of prescriptive decision theory against orthodox subjective and rationalist readings. I say doctors are right to prescribe drugs for me, not because I think (or have reason to think) they will work, but because they probably will work. I argue therefore that one should only maximise objective expected utilities, ie. expected utilities incorporating chances rather than merely subjective or epistemic probabilities. I defend this view against widespread ontological and epistemic objections, using arguments from chapters 6, 14 and 15 to defend objective utilities and chances, and to show that they are in fact easier to know than subjective utilities and probabilities1 Analytic philosophy and the self 1 Introduction Bishop Berkeley said in 1710, in the introduction to The Principles of Human Knowledge, ‘Upon the whole, | am inclined to think that the far greater part, if not all, of those difficulties which have hitherto amused philosophers, and blocked up the way to knowledge, are entirely owing to ourselves — that we have first raised a dust and then complain we cannot see’ (Berkeley 1710 p. 46). Those remarks of Berkeley’s seem to me just as true now as they were in 1710. Indeed matters are in some ways worse now than they were then. For one thing, philosophers today are too rarely amused by the difficulties that block the way to knowledge. They should be amused, because philosophy has to deal amongst other things with the limits of what makes sense: that is, with the boundary between sense and nonsense, which is the very stuff of humour. Take this example from Lewis Carroll’s Through the Looking-Glass: ‘Who did you pass on the road?” the King went on, holding out his hand to the Messenger for some more hay. ‘Nobody’, said the Messenger. “Quite right’, said the King: ‘this young lady saw him too. So of course Nobody walks slower than you. * ‘I do my best’, the Messenger said in a sullen tone. ‘I’m sure nobody walks much faster than I do!” ‘He can’t do that’, said the King, ‘or else he’d have been here first.’ (Carroll 1887 pp. 85-6) It takes a philosopher to see why this is funny, to see why it’s nonsense to talk of Nobody as if he and she (Nobody is both male and female ...) were a being of some kind. The reason is, of course, that although the word ‘Nobody’ looks like the name of some being, it really isn’t a name at all: it’s a way of saying that there was not a being who walked either slower or faster than the Messenger. Now that’s a pretty trivial piece of philosophical analysis, which anyone could do: but as we shall see, there is much more serious (and more2 Analytic philosophy and the self misleading) nonsense than Lewis Carroll’s around, which it takes rather more analysis to expose and explain. To expose nonsense, however, we must first detect it: we need a nose for nonsense. And, as Ramsey said of Wittgenstein’s proposition that philosophy itself is nonsense, ‘we must then take seriously that it is nonsense, and not pretend, as Wittgenstein does, that it is important nonsense’ (Ramsey 1929c p. 1). Now I don’t think philosophy is nonsense, but I do think it includes taking the fact of nonsense seriously and saying why it’s nonsense. To do that, however, we need to be suitably amused by jokes like the one about Nobody, and to distinguish taking them seriously from pretending they’re important. But not all philosophers are suitably amused. Some I fear lack the serious sense of humour, and with it the nose for nonsense, that good philosophy needs. And that is a serious defect. For without a nose for nonsense, philoso- phers run a real risk of talking nonsense themselves, and (unlike Lewis Carroll) of persuading themselves and others that it’s important nonsense. None of this would matter much if philosophy were read and judged only by other philosophers, as mathematics is by mathematicians, who can, on the whole, tell when their colleagues are talking nonsense. But it isn’t, even though perhaps it should be, since philosophy is really no more of a spectator sport than mathematics is — by which I mean that it’s not like poetry, for example, which you needn't be a poet to judge, whereas you do need to be a philosopher to judge philosophy, just as you need to be a mathematician to judge mathematics. Of course philosophy, like mathematics, is also read by outsiders who don’t want to judge it, but rather to take it on trust and use it, just as physicists use mathematics. But not many outsiders want philosophy to do physics with: on the whole, they want it to provide a kind of secular substi- tute for religion. In other words, they want their philosophers to be gurus. And the last thing disciples want in gurus is a sense of humour: it’s inimical to the air of authority which attracts disciples to gurus in the first place. So when philosophical gurus raise a dust by talking important-sounding nonsense, their disciples, far from complaining that they cannot see, are all the more impressed by the profound obscurity of the proffered view. In philosophy, therefore, as in religion and medicine, a gullible public will often give much fame and fortune to mystery-mongers. What has all this to do with analytic philosophy? Well, to pursue Berkeley’s metaphor, philosophical analysis is, as even my trivial example of it illustrates, a kind of intellectual sprinkler system, whose function is to lay the conceptual dust which obscures our view of the world. This indeed is one of its primary objects: to detect and dissipate the bogus mysteries which nonsense generates,Analytic philosophy and the self 3 like Lewis Carroll's little mystery about Nobody, so that the world’s real mysteries can be more clearly seen and thereby — we hope — better appreciated and understood. In this sense, good philosophy has always been analytic. Analysis is more a matter of technique than of doctrine, and it is as evident in Plato, Aristotle, Aquinas, Leibniz, Hume, Kant and Mill as in any modern analytic philosopher. What if anything distinguishes analytic philosophy, so-called, is that it not only uses analytic techniques, but is explicitly concerned to develop and assess them: not of course as ends in themselves, but as means to philosophical understanding. But not of course the only means, since an analyst always needs non-analytic material to analyse. Analysis can no more provide a complete philosophy on its own than — for example — democracy can provide a complete politics: because, obviously, accepting the principle of majority rule doesn’t tell you whom or what to vote for, or why. No political democrat, in other words, can just be a democrat; and in much the same way, no philosophical analyst can just be an analyst. Which is not of course to deny that analysis matters, just as democracy matters; nor to deny that it can conflict with philo- sophical nonsense (like the being of Nobody), just as democracy can conflict with political nonsense (like the one-party state). But whereas everyone can feel that democracy matters, and can more or less see why, it is less obvious to non-philosophers why philosophical analysis matters. If philosophy in general is not really a spectator sport, what can analytic philosophy, in particular, offer to the rest of society? Well, I could say, for a start, that it offers, because it demands and encourages, a socially desirable temperament. A nose for nonsense isn’t only an asset in philosophy. A sense of humour, and hence of proportion, is a powerful antidote to politi- cal and religious fanaticism. An insistence on explicit discursive understanding where it can be had, as opposed to obscure intimations of ineffable insight, is a great deterrent to charlatanism of all kinds. A commitment to truth, and hence to basing one’s beliefs on evidence rather than on wishful thinking (however high-minded), is essential not only to good science, but to all serious attempts to acquire knowledge and understanding about anything, including ourselves. ‘And the feeling for reason which analysis gratifies helps to combat a recurrent tendency to elevate feeling at the expense of reason, as if they were opposed, and as if we didn’t need both. Society, however, is not only indebted to the temperament that analytic philosophy fosters. The results of analysis too have had many uses outside philosophy itself, although I don’t wish to exaggerate them, or to accept that they provide its main justification: philosophy, like mathematics, has a value of4 Analytic philosophy and the self its own, independent of its applications. Still, those applications are remark- able enough: ranging from the invention of computers (based on analyses of the concepts of mathematical proof and truth) to debates on abortion, which turn on concepts of life and of humanity whose analysis is far too important to be left to people with particular religious (or anti-religious) axes to grind. But besides all this, I think analytic philosophy serves society most distinc- tively when it increases our understanding by clarifying concepts that concern everyone, whether they are philosophers or not. And in the rest of this lecture T should like to illustrate how it can do that by discussing one such concept, the concept of the self.! 2 The self We are all selves: that is, we are not just objects in the world, like sticks and stones, but also subjects, who experience the world, including ourselves and others; who learn about it, try to understand it, and value or disvalue various aspects of it; and who interact with it, both individually and collectively, because we want things to happen in it and to it, and in particular in and to ourselves and other selves. Being a self is obviously the most important, or at least the most interest- ing, fact about each of us; and to understand ourselves we must understand that fact. What is it to be a self, and in particular, what is it to be oneself: what distinguishes me from the rest of the world, including other selves, which I experience and with which I interact? Those are questions which concern all reflective people, not just philosophers: we all want to know and understand what makes us the selves we are. And answering those questions is a perennial project for all schools of philosophy. The project is perennial because our understanding of the self is inevitably affected by developments not only in philosophy but in society, in religion, in morality and in the sciences, natural, psychological and social. But it is the job of philosophy, above all, continually to assimilate and assess the significance of such developments for our concep- tion of our selves. On this occasion I can discuss only a small part of this perennial project. But it is an important part, and one that I think shows both how easy it is to raise a conceptual dust which effectively blocks our view, but also how such dust can be laid by a sprinkling of analysis. The part of the project which I have in mind is the part that deals with self-knowledge, by which I mean first- !Versions of what follows have been discussed in lectures and seminars at the Universities of Cambridge, Edinburgh, Sheffield, Perth, Belgrade, Leuven and Nottingham, and at the 1989 Annual Conference of the Australasian Association of Philosophy in Canberra. I have been much helped in revising it by comments made on those occasions and also (in my discussion of the ‘knowledge argument’ in 3) by an unpublished paper of Paul Teller's.Analytic philosophy and the self 5 person knowledge, knowledge about oneself. This includes of course our first- hand knowledge of our own experiences, and in particular of what those experiences are like, a peculiar kind of knowledge that has itself attracted recent analytic attention (see e.g. Jackson 1986; Lewis 1989). But I want to discuss something more general, an aspect of all first-person knowledge, not only of our own experiences, but also of who we ourselves are, and of how the world is, or seems to be, from our own ‘perspective’ or ‘point of view’ — to use terms made fashionable recently by Thomas Nagel (e.g. 1983, 1986). First-person knowledge is commonly and naturally called ‘subjective’ because, as we shall see, it is, in an important sense, relative to the subject: that is, to the person who has the knowledge. But it is also important to realise that, in another equally common and legitimate sense of ‘subjective’, this kind of knowledge needn’t be subjective at all. This is the sense in which to be subjective is to be a mere matter of opinion, with any opinion (that is, belief) on the matter in hand as good as any other, because there’s no fact of that matter to make some one belief true and contrary ones false. Now in this sense, obviously, knowledge cannot be subjective: one can’t know something unless there is something to know, some fact of the matter which does make some belief about it true and contrary ones false. And that is the sense, incompatible with knowledge, in which some philosophers hold, for example, that moral and aesthetic values are subjective. They hold, in other words, that no belief about such values (except this one!) is made true by any fact: because, as a matter of fact, the world contains no objective values, and hence no moral knowledge. Well, maybe our moral and aesthetic beliefs are subjective in this sense, in which ‘subjective knowledge’ is a contradiction in terms. I don’t think so, but I'm not going to press that point here. The point | am going to press is that most, if not all, of our non-evaluative first-person beliefs are clearly not subjective in this sense. Take my beliefs that I’m Hugh Mellor and that I live in Cambridge. Those are by no means mere matters of opinion, with one opinion as good as any other. For the fact is that I am Hugh Mellor, and I do live in Cambridge, and any contrary beliefs, say that I’m the Pope and live in Rome, would be as false as the belief that the Earth is flat. These beliefs of mine, about who I am and where I live, are as objectively true as any beliefs can be. Moreover, whatever it takes to make true beliefs knowledge, no one (other than professional philosophical sceptics) would deny that these beliefs of mine qualify. Not only am I Hugh Mellor, I know I am; and not only do I live in Cambridge, I know I do.6 Analytic philosophy and the self So first-person knowledge, knowledge about myself, isn’t subjective in the sense of being mere opinion, with no facts to make it true. But it is subjective in the sense of being relative to the subject, to the person whose knowledge it is. It is, to use Nagel’s terms, ‘perspectival’ knowledge: knowledge from the knowing subject’s ‘point of view’. But these expressions — ‘perspective’ and ‘point of view’ — are as misleading as the word ‘subjective’ is, and for the same reason: because they too suggest that knowledge of this kind is subjective in my other sense. For what the phrases ‘from my perspective’ and ‘from my point of view’ really mean is ‘as I see it’ or ‘in my opinion’. And adding either of those phrases to an assertion either adds nothing — what, after all, is any sincere assertion but a giving of one’s opinion? — or it appears to admit that contrary opinions could be just as good. But in examples like the ones I've just given, I admit no such thing. Who I am, and where I live, are not mere matters on which one opinion is as good as any other. They are matters of plain, observable fact, as knowable to me as the shape of the Earth, or the fact that 2+2=4. What makes first-person knowledge peculiar, and problematic, is not, as the expressions I’ve just jibbed at suggest, that it’s subjective in the sense that would prevent it being knowledge at all. The peculiar problem it presents is this. When I know that I live in Cambridge, I know something which no one else can know. I don’t mean that no one else can know that Hugh Mellor lives in Cambridge: many people know that. Nor do I mean that, even if you knew that Hugh Mellor lived in Cambridge, you might not know that I do, because you might not know that I am Hugh Mellor. That possibility just exemplifies a well-known feature of knowledge — its non-extensionality — which is indeed a problematic feature, but not the one I want to discuss. The problem I want to discuss shows up not in the fact that you might not know that I live in Cambridge, but in the fact that you can’t know this, not in the first-person form in which I know it. In other words, you can’t know the first-person fact that appears to be what makes this belief of mine, that I live in Cambridge, true. The reason you can’t know this is that there’s no first- person belief of yours which this first-person fact could make true, not even if you live in Cambridge too. For even if you do, the fact that J live in Cambridge isn’t what makes your first-person belief, that you live there, true. The only belief of yours which the fact that I live there could make true is your belief that / live there. But for you that belief is a second-person or third-person belief, not a first-person one. So the fact which makes it true for you can’t be a first-person fact, and a fortiori it can’t be the first-person fact that I live in Cambridge. And if that fact doesn’t make that belief of yoursAnalytic philosophy and the self 7 true, it certainly can’t make any other belief of yours true. But what this means is that, for you, the first-person fact that I live in Cambridge isn’t a fact at all. The fact for you is the corresponding second-person or third-person fact: closely related to my first-person fact, but not the same. In short, the first-person fact that I live in Cambridge is only a fact for me. For anyone else, it isn’t a fact at all. And that’s why you can’t know it: because for you, there’s no such fact to know. First-person knowledge seems therefore to be subjective in the sense not only that the knowledge is relative to the subject, but that the facts which are thereby known are relative to the subject. And that appears to present a serious ontological problem, whose solution could profoundly affect our conception of ourselves and of the world we live in. The problem, I must emphasise, only arises because this first-person knowledge really is knowledge: that is, because it isn’t just a matter of opinion. Moreover, whatever facts make my first-person beliefs about who I am and where I live true clearly don’t depend on my knowing or believing them. They would still be facts even if I lost my memory, and had no idea of who I was or where I lived. And as in these examples, so in general. First- person facts are not generally reducible to first-person knowledge or belief: they are not subjective in that sense either. Whether I know it or believe it or not, it is a fact for me, if only for me, that I am Hugh Mellor and that I live in Cambridge. What makes my knowledge of these facts subjective, therefore, is not just that I have some kind of epistemically privileged access to them which other people lack. That may be true of Hugh Mellor’s objective knowledge of Hugh Mellor’s experiences. But my first-person knowledge, whether of my experiences or of who I am or where I live, is not just epistemically restricted to me. It isn’t only knowledge of these facts that’s confined to one person, it’s the facts themselves. And this is what poses the ontological problem. For if there really are subjective facts, which are not facts for everyone, but only for the subject, then since there are many different subjects, there are correspondingly many different worlds. My world contains the subjective facts that I am Hugh Mellor, and that I live in Cambridge. Your worlds contain neither of those facts, not even if it’s a fact in your world that you live in Cambridge too. So it seems that we don’t in fact live in a single objective world. Our worlds have an objective overlap, of course: they all contain the objective facts that Hugh Mellor lives in Cambridge, and that the Pope lives in Rome. But the Pope’s world doesn’t contain the subjective fact that I live in Cambridge, or that I’m8 Analytic philosophy and the self Hugh Mellor; any more than my world contains the subjective fact that I’m the Pope, or live in Rome. The world of objective facts — the facts which, being facts for everyone, are common to all our subjective worlds — seems therefore to be only a small part of any of our worlds. And not the most important part either. For although my subjective facts include some very trivial ones (like how many eyelashes I have), they also include all the facts that are most important to me: who I am, whom I know and am related to, when I live, and where; how old I am; what my experiences, abilities, beliefs, desires, values and projects have been, are and will be. And all these fundamental facts about me lie completely beyond the reach of any objective science, whether it be physics or psychol- ogy: for objective science, by mere definition, concerns itself with facts that are facts for everyone, which these are not. I shall call the account of first-person knowledge I’ve just given, which takes it to be knowledge of subjective first-person facts, ‘subjectivist’. The subjectivist account presents a picture of the world — or rather of many worlds, one for each of us — about whose most important factual aspects, including who, where, when and what we ourselves are, no objective science can ever tell us anything at all. That is a very striking metaphysical thesis, and if true, a very serious limitation on the factual scope of science. Not surprisingly, therefore, the subjectivist picture appeals most to those who are most resistant to the pretensions of science and, in particular, to its pretensions to encompass, if not all values, then at least all facts. While for the same reason, the picture appeals least to those who think the world just is the one objective world of physics and the other public sciences. Those who think that will tend therefore to dismiss these subjective facts just because they are subjective in the present sense. Russell (1940 p. 108), for example, claims that what are now called ‘indexicals’ — words like ‘I’, ‘now’, ‘here’ and ‘this’ — ‘are not needed in any part of the description of the world, whether physical or psychological’ (see also Quine 1954). In other words, there are no first-person facts: which seems to make my first-person beliefs subjective in the other sense of being mere matters of opinion, with any one opinion as factually good as any other. But that is incredible: as Nagel (1986 p. 57) says, ‘if it is not a fact about the ... world that I am [Thomas Nagel], then something must be said about what else it is, for it seems not only true but ... one of the most funda- mental things I can say about the world’. And even if we lack Nagel’s sense of self-importance, we may still agree with him, against Russell, that ‘it provides a clear example of the ineliminability of indexicals from a complete concep- tion of the world’.Analytic philosophy and the self 9 Whom then shall we follow? The subjectivist, who gives us our first-person facts and selves, but traps us in our own subjective worlds? Or the objectivist, who puts us all together in a single world but denies the objective truth of all our first-person beliefs? Anyone with a nose for nonsense will surely say: neither. There must be a more sensible way to go; and there is. But to see it we need a sprinkling of analysis to lay the conceptual dust that has so far obscured our view of it. 3 Subjective truths, objective facts The trick is to distinguish more carefully than I have so far done between truths and facts. So far I have simply assumed that true beliefs are made true by facts which correspond to them. This doesn’t, I should say, commit me to the controversial correspondence conception of truth, because I am not using this concept of correspondence to define truth. If anything, I would use it to define facts (as whatever makes beliefs true), leaving truth to be defined in some other way, which I needn’t discuss here. What matters here is not what, if anything, such a concept of correspondence can define, but simply that, if truths are to correspond to facts at all, they must do so in a more complicated way than I have so far implied. Take Lewis Carroll’s Messenger, who believes he passed nobody on the road, and suppose this belief of his is true. As I’ve already said, what makes it true is not the apparently corresponding fact that he passed some being, namely Nobody. There’s no such fact, because there’s no such being. There can’t be, as we can see by supposing that it’s also true that nobody passed the Messenger. But no being can both pass and be passed by the same person at the same time: that’s nonsense, which is the whole point of Lewis Carroll's joke. This particular nonsense, as we've seen, has a very simple explanation. The fact which really makes the Messenger’s belief true is the fact that there was no being that he passed: a fact which is perfectly consistent with there also being no being who passed him. End of analysis: dust laid. And now we can see quite clearly how beliefs which are apparently about Nobody are really made true by facts which incorporate no such impossible being. It is perhaps less obvious how first-person beliefs can be made true by facts that incorporate no first-person self, no /. But they can. Alll first-person beliefs can be made true by purely objective facts. Take my belief that I live in Cambridge. That obviously corresponds to, and can therefore be made true by, the fact that Hugh Mellor lives in Cambridge, because I am Hugh Mellor. And this true belief (that I’m Hugh Mellor) can in turn be made true by the10 Analytic philosophy and the self fact that it’s held by Hugh Mellor, a fact which likewise contains no first- person constituent. Now if all true first-person beliefs can be made true in this way by objec- tive facts, we can escape the dilemma I presented earlier. We can have real first-person knowledge in a single, wholly objective world. And if there are facts about that world which our public sciences can’t discover, it will only be because they’re practically or physically impossible to discover, not because they’re first-person and therefore not facts for everyone. But can this be so? Can all true first-person beliefs really be made true by objective facts? To show that they can, we must first produce enough objective facts to correspond to all our true first-person beliefs. Now that, oddly enough, is both quite easy to do and not especially controversial. No one denies that the truth of my belief that I live in Cambridge does at least correspond to the fact that this belief is held by someone who lives in Cambridge. And as for me, so for everyone. Any belief of the form ‘I live in town T° will be true if and only if, as a matter of fact, it is held by someone who lives in town 7. No one disputes this (see e.g. Nagel 1983 p. 216). Nor does anyone dispute the objectivity of these corresponding facts. Everyone accepts that it’s a fact for everyone, and not just for me, that my beliefs are held by Hugh Mellor, who lives in Cambridge. And similarly for all first-person beliefs — including, incidentally, beliefs about one’s own experiences. The truth of any belief whose content is ‘I’m in pain’ will always correspond to the objective fact that its owner is, at that very time, in pain. None of this is seriously denied. But why then would anyone deny that these objective facts are what make our first-person beliefs true? Why should anyone continue to believe in such mysterious first-person facts as my being Hugh Mellor and my living in Cambridge when they are as clearly redundant as they are ontologically problematic? The main reason is the following so-called ‘knowledge argument’. Take my knowledge that I live in Cambridge. | say that the belief which constitutes this knowledge is made true by the objective fact that the believer, Hugh Mellor, lives in Cambridge. But that isn’t what I know by knowing that / live in Cambridge. For suppose I forget who I am, or go mad and believe that I’m the Pope and that Hugh Mellor - whoever he may be — doesn’t live in Cambridge at all. I can still know that / do. But that means that I can know this without knowing, or even believing, the very fact — that Hugh Mellor lives in Cambridge — which I say is what makes my belief about where I live true.Analytic philosophy and the self 11 And even if I do know that Hugh Mellor lives in Cambridge, that will have nothing to do with my knowing that I do if I don’t believe I’m him. Knowing that I live in Cambridge cannot therefore be equated with know- ing the objective facts which I say make that belief of mine true, since I can easily know the one without knowing the other. And as for this piece of first- person knowledge, so, obviously, for all others. But then first-person knowledge looks after all as if it must be what it seems to be, namely knowledge of subjective first-person facts, which must therefore exist: since one can’t very well know what isn’t the case. What’s wrong with this argument? As an argument for a difference between subjective and objective beliefs, nothing. My first-person belief that I live in Cambridge must indeed differ from my objective belief that Hugh Mellor does so. But this, as we shall now see, doesn’t mean that my first- person belief must be made true by a first-person fact. To show why not, I must first draw another distinction: between belief types and tokens of those types. A belief type is defined by a belief’s content (what is believed), regardless of who believes it or when. A token of a belief type is a particular person holding a belief with that content at a particular time. (Take for example the belief that we’re human: one belief type with, at any instant, millions of tokens. ) Now what the knowledge argument I’ve just given shows is that my belief that I live in Cambridge differs in content from my belief that Hugh Mellor does so, even though I am Hugh Mellor. How is this supposed to show that my first-person belief must be made true by a first-person fact? Well, suppose we accept the idea that the content of a belief type is its ‘truth condition’: that is, the condition in which it’s true. (This idea is controversial in various ways, but none of them affects the present argument. ) For example, the content of our belief that we’re human is obviously the condition in which that belief is true, namely that we are indeed human. But then mustn’t the first-person content of my belief that I live in Cambridge also be its truth condition? In other words, mustn’t the first-person fact that I live in Cambridge be what makes this first-person belief true? Not at all. All that follows from my assumptions is that the truth conditions of first-person belief types must differ from those of objective belief types. But this doesn’t mean that the former can’t be objective: they can be; and I say they are. The only difference between the two kinds of truth conditions is that the first varies from person to person and the second doesn’t. For example, the objective truth condition of the objective belief that Hugh Mellor lives in Cambridge, namely that Hugh Mellor lives in Cambridge, is the same for12. Analytic philosophy and the self everyone. Whereas the objective truth condition of anyone’s token of the first- person belief ‘I live in Cambridge’ is that that person lives in Cambridge. In other words, we can identify the content of the belief type ‘I live in Cambridge’ with a function (in the mathematical sense) from any person who has this belief to the objective truth condition of his or her token of it: namely, that this person lives in Cambridge (see Kaplan 1979). And this is a completely objective function: from objective people, like Hugh Mellor or the Pope, to objective truth conditions, like Hugh Mellor living in Cambridge or the Pope doing so. It requires the existence of neither first-person selves nor first- person facts. In particular, it lets my belief that I live in Cambridge be made true by the objective fact that Hugh Mellor lives there. But because it doesn’t identify the content of my belief with that fact, it needn’t identify the belief itself with the belief that Hugh Mellor lives in Cambridge. In short, it simply doesn’t follow, from the evidently different contents of subjective and objective beliefs, that the subjective ones are made true by subjective rather than by objective facts. As an argument for first-person facts and selves, the knowledge argument is a simple non sequitur. Moreover, this objective account of how first-person beliefs differ from objective ones shows how I may not believe that I’m Hugh Mellor even though I am Hugh Mellor and even though I obviously do believe that I’m me and that Hugh Mellor is Hugh Mellor. The reason of course is that these three belief types have different contents: that is, different functions from any person P who has those beliefs to the truth conditions of his or her tokens of them. The content of ‘I’m me’ is the function from P to P’s being P, and that of ‘Hugh Mellor is Hugh Mellor’ is the constant function from P to Hugh Mellor’s being Hugh Mellor, neither of which anyone can doubt. That doesn’t prevent anyone, even Hugh Mellor, doubting the content of ‘I’m Hugh Mellor’, namely the function from P to the truth condition that P is Hugh Mellor — a function which incidentally shows at once why only I, Hugh Mellor, can know that I’m Hugh Mellor, since only Hugh Mellor’s tokens of this belief type can satisfy its truth condition. This account thus immediately explains, and explains away, the idea of subjective first-person facts. They have been invalidly inferred from the way in which the truth of first-person beliefs depends on who holds them; just as Lewis Carroll’s Nobody is invalidly inferred from the way in which the truth of the Messenger’s belief depends on his not having passed anyone. Equally importantly, this account also explains, as we've just seen, why my knowing who I am is not an automatic and therefore trivial consequence of my being who I am. And finally, it shows clearly how knowing who I am differs fromAnalytic philosophy and the self 13 knowing any objective facts, even if those are all the facts there are. It therefore shows how, even in a single objective world, knowledge of one’s own identity can be a real and irreducibly extra piece of knowledge, which may matter to people as much as it does to Nagel or as little as it does to me. And this in turn enables this account to reconcile Russell’s claim that we need no indexicals to describe the world with Nagel’s claim that we do need them to express our first-person conception of it. In short, this objective account of the content of first-person beliefs lays all the conceptual dust which subjectivism raises about the nature of first-person knowledge in general, and knowledge of one’s own identity in particular. And dissipating the apparent mystery about the first-person self in this way enables us to see more clearly the real problems which our concept of the self presents: namely, how do selves differ from sticks and stones and computers and lower forms of life? In other words, what does it take to be a self, capable of agency, consciousness, knowledge — including first-person knowledge — and understanding? That is where the real mysteries lie, which we will understand much better and faster once we distinguish them from the bogus mystery of what it takes to be shis self. Once we know what a self is, there need be no more mystery about that than there is about what it takes to be this room when we know what a room is. 4 Subjective selves This however is not quite the end of the matter. So far I have shown only that first-person knowledge needn't be knowledge of first- person facts, not that it can’t be. To show that, we must look more closely at what follows if it is: namely, that there are such facts. So let’s suppose, for the sake of the argument, that there are first-person facts, and that my living in Cambridge is one of them, and hence different from the fact that Hugh Mellor lives there. How can these two facts differ? Well, obviously, they can differ only if I differ from Hugh Mellor: that is, only if I, my first-person self, and Hugh Mellor are distinct beings. And so perhaps we should be, from a subjectivist’s point of view. For one thing, I, as a constituent of all my first-person facts, exist only in my own subjective world; whereas Hugh Mellor, as a constituent of many objective facts, exists in everyone’s world. How then can he and I be identical? And if we aren’t identical, then we do have (both of us!) a very simple explanation of how I can know first-person facts about myself without knowing the corresponding objective facts about Hugh Mellor: namely, that they are quite different facts, about quite different beings. And that includes, not just external facts about where I live, but internal facts about my experi-14 Analytic philosophy and the self ences, beliefs, desires, hopes, fears, emotions — in short, all the first-person facts which constitute my inner life. All this makes me, my first-person self, as opposed to Hugh Mellor, look rather like the immaterial ego of Descartes: a ‘thinking thing’ defined by its introspectible attributes, and distinct from the spatially extended body which it inhabits, and whose demise it could conceivably survive. In fact there are important differences, both between my first-person self and a Cartesian ego, and between Hugh Mellor and my body. For one thing my first-person self, unlike my Cartesian ego, is just as embodied and located in space as Hugh Mellor is; and for another, Hugh Mellor, unlike my Cartesian body, thinks just as much as I do. But as these differences, though striking, don’t really matter for present purposes, I propose, for convenience, to call the subjectivism which denies my identity with Hugh Mellor ‘Cartesian’. Cartesian subjectivism, like Descartes himself, can conceive of me, my first-person self, surviving the death of the objective Hugh Mellor, who lives in Cambridge, and to whose body I am now, for better or worse, so literally attached. That is, I now get all his (and only his) hangovers and other bodily pains, and pleasures; I see only what falls within his field of view; I hear only what strikes his ear; I touch only what he reaches; and so on. But when Hugh Mellor dies ... ah, then my life could be something else again, whether reincarnated in this terrestrial world, or freed of all bodily constraints in some celestial one. Cartesian subjectivism can appeal to all these more or less enticing Cartesian possibilities, but only at the cost of inheriting all the many well- known problems that face Descartes’ conception of the self. How for example does my first-person self interact causally with Hugh Mellor’s objective body, as it clearly does, since what hits him causes me to get hurt, and what I decide to do causes him to do it? Or again, what constitutes my self’s identity through time, if not the causal continuity of Hugh Mellor’s body? And so on. Those are indeed serious problems, and consequently serious objections to Cartesian subjectivism. But they aren’t the decisive ones. It’s not the causal links between Hugh Mellor and me that really reveal the nonsense in Cartesian subjectivism: it’s the logical links. The first thing to notice is that subjectivists can only take this Cartesian way out by denying the very first-person fact that gives subjectivism its appeal in the first place: namely the fact that I am Hugh Mellor. Now I too, of course, deny that this identity is a fact. I say there are no such first-person facts of identity: there is just Hugh Mellor. But I don’t thereby deny the truth of my belief that I’m Hugh Mellor. On the contrary, as we’ve seen, I can admit andAnalytic philosophy and the self 15 account quite easily for the truth of this belief, and give it a content which only I can know, without appealing to any first-person facts at all. Whereas a Cartesian subjectivist who denies the fact that I’m Hugh Mellor must also deny the truth of my belief that I’m Hugh Mellor: a denial which I must say I find completely incredible. Next, by making me a being distinct from Hugh Mellor, the Cartesian subjectivist reduces to causal links what are obviously logical ones. It is, for example, quite obvious that by living in Cambridge, Hugh Mellor doesn’t j cause me to live there too. His doing so entails my doing so: because I am him. Similarly, when he takes the train to London, he doesn’t just drag me along causally, a first-person free rider on his objective ticket, two passengers for the price of one. I have to go with him, logically have to, because I am him. British Rail are quite right to issue only one ticket; and when we (Hugh Mellor and I) travel on it, we aren’t swindling them at all, because there really is only one of us. At least that is what I say, and I’m an objectivist. And if I can say it ~and say it truly — then a subjectivist had better be able to say it truly too. In short, Cartesian subjectivism makes no sense. If there are first-person facts, then the fact that I’m Hugh Mellor must be one of them. But admitting that fact just impales our subjectivists on the other horn of a dilemma. For if I really am Hugh Mellor, then my living in Cambridge, travelling to London, having various experiences, and so on, cannot be different facts from Hugh Mellor’s doing and experiencing those very same things. But then since those facts about him are all objective, the same facts about me must also be objective. In other words, our first-person beliefs can no more be about first- person selves, constituents of whole arrays of first-person facts, than the Messenger’s beliefs can be about Nobody. So not only can first-person beliefs be made true by objective facts, in the way I’ve outlined, they must be. Our first-person beliefs can only refer to the objective people, the ordinary selves, whose beliefs they are, and who are constituents of nothing but objective facts. (For a detailed account of how people’s first-person beliefs refer to them, see chapter 2.) I conclude therefore that the subjective mystery of the first-person self not only can be dissolved, it must be. Like Lewis Carroll’s Nobody, the idea of the first-person self is nonsense; and so, as Ramsey said, we must take seriously that it is nonsense, and not pretend, as Nagel does, that it is important nonsense. Our being subjects does not make us subjective beings, trapped in our own subjective worlds. No such worlds exist. There is just one objective world, a world of which each of us, every self, every subject, is equally — and wholly —a part.16 Analytic philosophy and the self To reject the first-person self in this way is, however, as I have tried to show, by no means to deny that we all have our own different first-person knowledge of ourselves, and of how things are, or seem to be, from our own point of view. Nor is it to deny that we have this knowledge by having true first-person beliefs: beliefs whose contents are quite different from those of any objective beliefs. For there is, as we’ve seen, no objective mystery about these differences: they follow simply and solely from the way in which the truth of first-person beliefs depends on whose beliefs they are, which the truth of objective beliefs does not. That is why no first-person facts or selves are needed to enable us all to have our very different but equally true beliefs about who we are, where we live, and what we want, experience and are doing. This manifest diversity in our first-person knowledge of ourselves and of other things shows neither that we all live in different worlds, nor that our supposed knowledge is really mere opinion, corresponding to no objective facts. That is a completely false dichotomy. The differences in our true first-person beliefs are simply conse- quences of our being objectively different people, with different objective histories and experiences, and differing in our objective relations to the rest of the one objective world we all inhabit. Those objective differences are quite enough to enable all our different first-person beliefs to be equally true, thus enabling them to constitute the very different first-person knowledge which we all know that each of us has. End of analysis: dust laid.2 Iandnow 1 Introduction Many philosophers overrate the present subject. Pace Nagel (1983), there are no subjective facts or selves; nor, pace many others, does our ability to think and talk about our present selves, and the world as seen from our present point of view, pose any special metaphysical, semantic or epistemic problems. That’s the message of this paper. I start with some underrated truisms about the token-reflexive truth condi- tions of ‘subjective’ (by which I mean first person present tense) sentences. I then ask how ‘I’ and ‘now’ refer to their users and to when they’re used. To answer that question, I turn to the beliefs we use these words to express: what it takes to have them, to have them consciously and to put them into words. None of this involves subjective facts or selves, internal representations of oneself or the present, or any problematic capacity for, or concept of, self- reference or self-knowledge. 2 Subjective truths, objective facts ‘I’ and ‘now’ are often called ‘indexicals’. I shall call them ‘token-reflexives’, which describes their relevant feature: the way they make the truth of sentence tokens depend on facts about those tokens. Suppose some man, K, faces food at some time, T; and take the sentence type S(KT) _ K faces food at 7, tenselessly (i.e. as implying nothing about whether T is past, present or future)! unlike the semantically present tense S(KN) _ K faces food now, which implies that K’s facing food is temporally present. S(KT) is not token-reflexive: the truth of S(KT)’s tokens — s(KT)s — does not depend on when they occur. But that of S(KN)’s tokens does: an s(KN) is 1s T must be a date, like 1 January 1988, not a tense like now; K similarly is someone like Napoleon, not someone like /; and ‘K” and ‘7” mustn’t themselves be token-reflexive. a18 land now true if and only if it occurs while K faces food. S(KN) is temporally token- reflexive. Now consider the tenseless but first person sentence SUIT) I face food at 7, which says of whoever I am that I face food at T. So a token, s(IT), of SUIT) is true if and only if produced by someone who faces food at 7: its truth depends on who produces it. S(T) is personally token-reflexive? Finally, the subjective sentence SCIN) | face food now, which implies that I (whoever I am) face food now (whenever that is), is both temporally and personally token-reflexive: its tokens, s(/N)s, are true if and only if produced by people who are simultaneously facing food. What then are the truth conditions of s(KN)s, s(/T)s and s(IN)s? We know their truth depends on when or by whom they're produced. So either (1) their truth conditions vary from time to time or person to person or (2) whether those conditions obtain varies. The temporal alternatives are well known. (1) is the ‘tenseless’ view, on which any s(KN) is true if and only if K faces food at whatever time Y it occurs: a condition that varies with Y. (2) is the ‘tensed’ view, on which an 5(KN)'s truth condition is always that K faces food now. Only this condition doesn’t always obtain: sometimes K faces food now, sometimes he doesn’t. Which view is right? The tenseless view: because, as McTaggart (1908) showed long ago, there is really no such time as now. There can’t be. If there were, then since all sentence tokens of all the types ‘The time is now Y’ are true if and only if they occur at Y, every time Y would have to be both now (to make the true tokens true) and not now (to make the false ones false), which it can’t be (see McTaggart 1927 ch. 33; Mellor 1981 ch. 6). Reality can no more include a time (now) or a property (being now) satisfying this impossible condition than it can include a barber who shaves all and only those who don’t shave themselves. In short, though there are many tensed truths, there are no tensed facts: where by a ‘fact’ I mean a truth condition that obtains. Just as ‘Nobody shaves all and only those who don’t shave themselves’ can’t be made true by a fact involving a real Nobody, so no s(KN) can be made true by a fact involving a 2Not all personal (i.e. personally token-reflexive) sentences are first person, of course (e.g. “My mother faces food’, *You face food’, ‘He faces food’), just as not all tensed sentences are present tense. But the first person ones are all that need concern us here,Tandnow 19 real now. All s(KN)s, like all other tensed sentence tokens, have purely tense- less truth conditions. But what then do tensed sentence types and tokens mean?3 Their meanings can’t be identified with their truth conditions, since those vary from time to time, which their meanings don’t. And S(KN) never means what S(KT) means, not even at 7, when S(KN) and S(KT) have the same truth condition: that K faces food at T. So what does S(KN) mean? I say with Kaplan (1979) that S(KN)’s meaning is a semantic function, Sy), from the time Y when any s(KN) occurs to its truth condition: K facing food at Y. Of course fx,(Y)’s values vary with its argument Y, but fe(¥) doesn’t: nor therefore does S(KN)’s meaning. And its meaning always differs from S(KT)’s, even when Y=T. For S(KT) means a constant function, fur (= K facing food at 7), which S(KN) doesn’t. fgy(¥) and fxr are never the same function, even when Y=T and they have the same value: so even then, S(KN) and S(KT) differ in meaning, as they should. As for times and S(KN), so for people and S(IT). (1) is what I shali call the ‘impersonal’ analogue of the tenseless view, which says that an s(/7) is true if and only if produced by someone, X, who faces food at T: a condition that varies with X. Whereas on (2), the ‘personal’ view, every s(/7)’s truth condi- tion is the same, whoever produces it: namely, that (as its producer would say) I face food at T. Only this condition doesn’t obtain for every J: some Is face food at T, some don’t. Again (1) is right. The personal view can be disproved by an analogue of the argument against the tensed view. There can’t be personal truth conditions. If there were, then since all sentence tokens of the form ‘I am X’ are true if and only if produced by X, every person X would have to be both / and not /, which no one can be. Reality can no more contain a person (J) or a property (being me) satisfying such a condition than it can contain a time now or a property of being now. All s(/T)s, like all other personal sentence tokens, have purely impersonal truth conditions. Though there are many personal truths, there are no personal facts. But what then do S(/7) and its tokens mean? Again, they can’t mean their truth conditions, because those vary from person to person, which their meaning doesn’t. And no s(IT) ever means what an s(K7) means, even when they’re both produced by K and have the same truth condition: K facing food at T. So what does S(/T) mean? 3A sentence = type has has a meaning only if all its tokens have the same meaning. So I ascribe meanings indifferently to both.20 and now It means another semantic function, f;;(X), from whomever, X, produces a token s(/7) to its impersonal truth condition: X facing food at T. For again, although f;,(X)’s value varies with its argument X, f,,(X) doesn’t: nor therefore does S(/7)’s meaning. Nor does S(/T) ever mean what S(KT) means: having the same value for X=K doesn’t make f,7(X) the same function as fxn And as for tensed and personal sentences, so a fortiori for subjective ones. They can’t be made true by subjective facts, which would have to be both tensed and personal and can’t be either. So no s(IN) is ever made true for me, whoever I am, by the subjective fact that I face food now. There’s no such fact, because there’s no such person as / and no such time as now. An s(IN) can only be made true by the objective fact that whoever (X) produces it, at whatever time (Y), simultaneously faces food: a semantic function, f,(X,Y), which is indeed obviously just what S(IN) always means. 3 The referents of ‘I’ and ‘now’ So much for meaning. What about reference? The terms ‘K’ and ‘7” refer to the man K and the time T. That is, they make any s(K7) say that it’s K who faces food, and faces it at T. They make s(K7)’s truth depend on a fact about that man and that time. What do ‘I’ and ‘now’ refer to? What person and what time do they make part of an s(JN)’s truth condition? Not the person J or the time now, since (as we've seen) no such person or time exists. The type words ‘I’ and ‘now’ have no one referent. But their tokens do. They refer to those who produce them, and to when they’re produced: because they make each s(JN) say about its producer, and about when it’s produced, what every s(KT) says about K and T: namely, that that person faces food at that time. How are these references made? There’s a familiar answer for terms like ‘K’ and ‘7: their referents are functions of known ‘senses’, semantic properties given (say) by definite descriptions. ‘K’ and ‘T" refer to whatever man (K) and time (7) satisfy the descriptions that we would use to give the senses of those terms. I think this answer is inadequate even for ‘K’ and ‘T’, and it’s hopeless for ‘I’ and ‘now’. For since their referents vary from user to user and time to time, so must their senses. But how then do we know what they are? Not by knowing definite descriptions, since I could forget almost everything about myself and the present time and still know that / had done so now. Anyone, X, at any time, Y, can use ‘I’ to refer to X and ‘now’ to refer to Y, without knowing descriptive senses for these words remotely sufficient to distinguish X and Y from all other people and times.Tandnow 21 But perhaps tokens of ‘I’ and ‘now’ have senses that are ineffable, like tastes and smells? But that doesn’t explain how we know them, because unlike tastes and smells, senses aren’t self-intimating. I may have no sense at all of the time: my ‘now’ will still refer to it; nor of who I am: my ‘I’ will still refer to Hugh Mellor. How can these references be fixed by senses available only now, and to me, but of which even I need now know nothing? And not even self-intimating senses would make these references as infal- lible as they are. The ‘now’ in a token s(IN) can’t fail to refer to when it’s produced; nor can my ‘I’ tokens fail to refer to Hugh Mellor. Self-intimation isn’t that infallible. Even I can mistake the taste of Glenfiddich for that of Glenmorangie, and misapply their names accordingly. I can’t do that with ‘I’ and ‘now’. These and related problems of self-reference have no generally accepted solution. They have even made Anscombe (1975) and others (see Diamond and Teichman 1979) deny that ‘I’ refers at all. But ‘I’ and ‘now’ do refer — or rather their tokens do, since they make the truth conditions of token sentences include their producers and when they’re produced. But to see how, we must turn away from sentences like S(IN) to the beliefs those sentences express. 4 Belief, truth and utility I can believe that I face food now without say- ing so, and even without being able to say so: animals need no public language to believe IN. (They may need a ‘language of thought’ (Fodor 1975), but that’s another matter.) Knowing when to eat comes before knowing how to talk: language use is an advanced activity, not a basic one. Agents needn’t be able to say what they’re doing, let alone what beliefs and desires make them do it. Those beliefs and desires needn’t even be conscious: my belief that what I’m facing now is food can make me eat it absent-mindedly while my conscious mind is miles away. Agency entails neither linguistic ability nor conscious belief. In particular, it doesn’t entail self-conscious belief. So we mustn’t infer from our own self-conscious linguistic use of ‘I’ that animals need a self- conscious or linguistically expressible ‘concept of the self” to believe IN. They don’t. Having a concept of the self means having a conception of it: an idea of what distinguishes selves from stones, and oneself from other selves. But that means having beliefs of the form ‘I am F’, where F is something that selves are and stones aren’t. So we need some subjective beliefs before we can have a concept of the self. And even they won’t be the first subjective beliefs we acquire. For first, simply to survive, we must be able to believe truly that we now face food, or predators, or a mate. Those are the first subjective beliefs22 = Land now our animal ancestors must have evolved the ability to acquire when facing food, predators, etc. Concepts of the self (and of the present) are a much later luxury. But how can the above semantics for consciously used sentences like S(/N) be applied to beliefs that need be neither conscious nor linguistically expressed? First, we apply the type-token distinction thus: a token, B(P), of a belief type B(P) is a fact of the form: X believes P at time Y.4 Next, the content P of any such belief is the meaning of a sentence, S(P), that would express it: i.e. a semantic function (constant or otherwise) from X and Y to the truth conditions which a token S(P) would have if X produced one at Y. Thus for P=IN, the content of X’s belief B(IN) at any time Y is the meaning of SUN), namely the function f,(X,Y) from X and Y to the truth condition that X faces food at Y. But beliefs are distinguished one from another not only by their contents but by how they affect our behaviour. I may be able to believe JN without doing so consciously, or saying so. But not without being disposed to act in various ways: for example, to eat what I face when I’m hungry. Of course different desires will make this belief make me act differently. It will make me act one way when I want to eat food, and in other ways when I want to cook it, freeze it or throw it away. And a different belief about where the food is would make the same range of desires make me do a quite different range of things. In short, believing JV makes me embody a causal function from desires to actions, just as a desire for food makes me embody a causal function from beliefs to actions. In other words, beliefs and desires affect each other's causal powers: believing IN affects how hunger makes me act; hunger affects how believing JN makes me act. The beliefs and desires that have these effects are of course tokens, not types: e.g. K believing /N, and wanting food, at T. Now these tokens, I have said, are facts, corresponding to the truth of sentences: ‘K believes IN at T, ‘K is hungry at 7”. But Davidson (1967a) has notoriously argued that causes and effects must be events, not facts: particulars, corresponding to singular terms, not to sentences. But he is wrong, as I and others have argued (see chapter 12). Most causation connects facts, and is rightly reported by a connective: e.g. (C) ‘K eats what he faces at T because K gets hungry at 7.5 By unqualified ‘belief" hereafter I may mean either a type or a token (or both) according to context; similarly for ‘desire’, etc. S(i) Pace Davidson, although some causal contexts are opaque (*K is the F because L is the G’), (C) needn't be: (C) and *K is the F” can entail ‘the F eats the F gets hungry at T. Ditto for ‘Tis the U" (see chapter 12). (ii) Causes and effects can’t really be simultaneousTandnow 23 So this piece of causation is also a fact, corresponding to (C)’s truth, which can therefore also be caused, e.g. by the fact that, just before T, K comes to believe IN. But how is B(/N)’s causal function from desires to actions linked to its content, i.e. to its semantic function f,(X,Y)? Well, consider Ramsey's (1927 p. 46) chicken, which believes ‘a certain sort of caterpillar to be poisonous’, a belief that Ramsey equates with the chicken’s ‘abstain[ing] from eating such caterpillars on account of unpleasant experiences connected with them’. These actions, Ramsey says, are ‘such as to be useful if, and only if, the caterpillars were actually poisonous’. ‘Thus’, he continues, ‘any set of actions for whose utility P is a necessary and sufficient condition might be called a belief that P, and so would be true if P, i.e. if they are useful.” In other words, true beliefs make desires cause actions that succeed in achieving the desired end. And although a successful action can be caused by a false belief (as when I sweeten my coffee with saccharine, believing it’s sugar), that’s a fluke. Only truth will ensure success in every action that a belief would combine with some desire to cause. Often of course beliefs must conjoin to make a desire cause an action, whose success will then depend on the truth of the conjunction. I won’t go to a shop which I believe has something I want unless I also believe it’s open, and even then I won’t get what I want unless both beliefs are true. Neither truth is enough on its own. So we can’t equate a belief’s truth conditions with those in which every action it helps to cause succeeds. But we can if we restrict the actions to those caused just by it and some desire.® Then its truth conditions are what I shall call its ‘utility conditions’: those in which all such actions would achieve the desired end. Ramsey’s pragmatist equation, of a belief’s truth conditions with its utility conditions, is all we need. With it we can quickly show how simply subjective beliefs like B(/N) get their personally and temporally token-reflexive truth (Mellor 1981 ch. 9): K’s hunger at ‘can't make him start to eat until some time T’after 7. But since the denseness of time lets T-T be less than any finite amount, the difference is immaterial. SIt doesn’t matter that this restriction is rarely satisfied; nor that having any one belief B(P) entails having many others: b(P) may still be the only token that makes a given desire cause a given action. (Compare the many ineffective entailments of ‘there was a short-circuit’ in ‘There was a fire because there was a short-circuit.) For a full elaboration and defence of Ramsey's idea, see Whyte (1990, 1991).24 Land now .e. how simply a token B(IN), X believing IN at Y, refers to X 5 How subjective beliefs refer Take any man X at any time Y, suppose him hungry then, and suppose some token belief b(P) makes this hunger of his cause him to eat what he faces. What belief must this be: i.e. what is b(P)'s truth condition? Ramsey’s equation tells us: its utility condition, that in which the action it makes X’s hunger cause succeeds. But that action, X eating what he faces, will succeed (by assuaging X’s hunger) if and only if what X faces is food. That is b(P)’s utility and hence its truth condition. But what token of B(P) is this: i.e. who is believing P, and when? X, of course, at Y. Beliefs don’t affect the causal powers of desires at a distance. Only my beliefs affect how my desires make me act, and then only while I have them. So for b(P) to make X’s hunger at Y cause an action, it must be a belief that X himself has at ¥. And since desires likewise don’t cause actions at a distance, that action (the eating) must also be X’s and must start at ¥8 an action which will therefore succeed in assuaging the hunger that caused it if and only if X faces food at Y. So b(P)’s utility and hence truth condition is that whatever agent has this belief faces food when he, she or it has it. But that’s the truth condition of a token of the subjective belief B(IN). So that’s what b(P) is. In other words, the belief agents need, to make their hunger cause them to eat what they face, is BOUIN). And that, in general, is why we need subjective beliefs: only they will make our basic desires (for food etc.) cause us to act to satisfy those desires — actions on whose success our lives frequently depend. And the mechanism that links such a belief to the agent and the time it refers to is simplicity itself: the contiguity of cause and effect. A belief can be made subjective, by how it affects the effects of an agent’s desires, only if the same agent has it, and at the same time. That's what gives it its personally and temporally token-reflexive truth conditions. Causal contiguity is how a subjective belief refers to whomever has it, and when. This at once explains the otherwise puzzling features of subjective reference noted in 3. For in order to believe /N, I need only be disposed 7Said of a sentence or belief, ‘refers to X" means ‘has X in its truth condition’; said of a referring term or its mental analogue, it - like ‘represents X’ — means ‘makes X part of the truth condition of the sentence or belief containing it’. The denseness of causation prevents it ever being really immediate, but for present purposes we may ignore the bodily causal intermediaries between desires and actions, and the consoquent delays caused by our Sze and the finite speed of causal transmission (see Dennett ).Tandnow 25 (inter alia) to eat what I face if I feel hungry: a disposition which causal contiguity ensures that only my simultaneous hunger can provoke, and only into making me eat, and only hen. That’s what makes my belief refer to me and to when I have it. And that’s why I need no idea who I am or what the time is, no concept of the self or of the present, no explicit or implicit grasp of any ‘sense’ of ‘I’ or ‘now’, to fix the referents of my subjective beliefs: causal contiguity fixes them for me. And does so infallibly. My apparent belief in KT could always fail to refer to K: K might not exist, or might not be the man I took him for — my belief might be about someone else entirely. But subjective beliefs can’t fail to refer in either of those ways. For no such belief can exist unless its referents do: X can’t believe JN at Y unless X and Y exist. And causal contiguity means that this b(/N) can only make X’s desires at Y cause actions, and then only actions done by X, and at ¥. The actions may still fail, if X doesn’t in fact face food at ¥: BUIN)’s utility and hence its truth condition may still not obtain. But the actions can’t fail by being done by the wrong agent or at the wrong time. Causal contiguity sees to that: and thereby makes every subjective belief refer infallibly to whomever has it, and when. 6 Self-reference That’s how causal contiguity enables agents without conscious or linguistic powers of self-knowledge or self-reference, or concepts of the self or of the present, to refer infallibly to themselves and to when they do so. But that’s not all causal contiguity does for subjective reference. It also enables agents and times to be represented by themselves in the subjective beliefs that refer to them. To see how, consider how K is referred to in beliefs like B(KT) and B(KN). What makes K part of their truth conditions? Suppose for example I want to write to K. This makes me write on the envelope what I believe is K’s address (A). That action will succeed if and only if K really does live at A (KA): which is what makes his living there the utility and hence the truth condition of my belief b(KA). But how can what I write be affected by K living, miles away, at A? It can’t, directly: there’s no such action at a distance, and anyway K may not live there — my belief may be false. So what makes my desire to write to K make me write ‘A’ on the envelope must be some causal surrogate for K living at A. I must, when I write (at time Y), have some intrinsic property V, — chemical, electrical or whatever — which makes my relevant desires make me act in ways that will succeed if K lives at A. This token property v, (the fact that I am V,26 land now at Y) embodies the belief B(KA) in me, much as ‘KA’ embodies its content here. v, refers to K as ‘KA’ does: K is in its truth condition. But v, also has a causal structure, part of which refers to K as ‘K’ does. For suppose that at YI get a new token belief about K, say b(KN), embodied in another token intrin- sic property v, My belief b(KA) will naturally make this give me the further belief, b(AN), that the man who lives at A now faces food. But for my b(KA) and b(KN) to interact like this because they share the referent K, v, and v, must share an intrinsic causal component, vx (like the ‘K’ in ‘KA’ and ‘KN’: see Fodor and Pylyshyn 1988). And that makes vx, whatever it is, a causal surrogate for K in these beliefs: an internal representation of K, which refers to K just as tokens of ‘K’ do. This does not of course reduce mental representation to intrinsic properties like vg: because intrinsic properties don’t refer intrinsically to anything. vy doesn’t refer to K intrinsically: it refers to K only by linking certain beliefs of mine, which refer to K because of what they make my desires about K make me do. But the intrinsic properties on which those desires depend for their effects also don’t refer to K intrinsically. They refer to K only because of what they make my beliefs about K make me do. So the hard questions remain. How do impersonal and tenseless beliefs and desires refer to people like K and times like 7? What links our internal repre- sentations to the people, times and other things to which our beliefs refer? Those are the real problems of mental reference, which theories of it try with varying success to solve. But subjective beliefs like BUN) pose no such problems, because their referents need no causal surrogates. Causal contiguity makes the fact that any X believes IN at any time Y affect only what effects X’s desires would have: which automatically puts X into that belief’s utility and hence its truth condi- tion. And it likewise makes the same belief affect what effects X’s desires would have only at Y, which automatically puts Y too into its truth condition. Similarly for all X’s other subjective beliefs at Y: e.g. the belief, BUK), that he is K. But this belief will naturally make X’s b(/N) give him the further belief b(KN): that K now faces food. So X’s D(IK) and b(IN) must share intrinsic causal components, like vx in my b(KA) and (KN). But unlike vg, these components come free with these beliefs, i.e. with the facts that X at Y believes [K and IN. For causal contiguity allows a token b(IN) to make a token (IK) cause a token b(KN) only when all three tokens occur in the same agent at the same time (i.e. when the inference is valid). So X and Y are themselves the shared intrinsic causal components that make both b(/K) and b(IN) refer toTandnow 27 X and Y: in other words, they represent themselves. And as in b(/K) and CIN), so in all subjective beliefs. They need no causal surrogates, no internal representations of the agents and times they refer to. And that’s why they pose no problems of reference: the relation of reference in subjective beliefs is simply that of identity. 7 Conscious self-reference We have seen that agents need no concepts of the self or the present to have subjective beliefs, and those beliefs contain no internal representations of the agents and times they refer to. But these beliefs aren’t, or needn’t be, conscious. It remains to be shown that conscious subjec- tive beliefs need no such concepts or representations. To believe something consciously I must believe that I now believe it: since to believe something consciously is at least to be conscious of believing it. So even an objective belief like b(K7), when conscious, involves a subjective belief: namely, the belief that I now believe KT. This ‘second order’ belief, I argue in chapter 3, is all that conscious belief is. Certainly it’s all that could affect its content, and hence any concepts or representations it might need. Other contentious aspects of conscious belief — sensations, Humean feelings of conviction - we may therefore set aside. We get our second order beliefs, I maintain, from an inner sense that I call ‘insight’, which mostly makes us believe we believe things we do believe, and not things we don’t. Just as eyesight is how we know about the things we see, so insight is how we know about our own present beliefs. Insight is the perceptual mechanism of our privileged, if fallible, access to what we believe. And insight not only gives us our second order beliefs, it fixes their referents, just as eyesight does. When I see a star, for example, my belief (that it is a star) refers to the object which, via my eyesight, causes that belief. Similarly, my second order beliefs refer to the first order beliefs which, via my insight, cause them. But these second order beliefs are subjective: they refer to my own simultaneous first order beliefs. So those must be the beliefs that cause them. And so they will be, thanks to causal contiguity. Insight being an inner sense, unlike eyesight, causal contiguity will let it show me only what I believe; which in turn lets it, unlike eyesight, work almost instantly, thus showing me what I believe only while I believe it. Suppose then that an agent X not only believes JN at Y, but does so consciously. That is, his belief makes him believe that he has it. Since both these beliefs are subjective, neither represents the agent or the time it refers to internally: in both beliefs, that agent and that time represent themselves. And causal contiguity makes the agent and the time the same in both beliefs: so the28 Iandnow second order belief needs no internal representation of the owner or the time of the first order belief that it’s about. In short, even conscious subjective beliefs contain no internal representa- tion of the agents and times they refer to: a fact which incidentally explains Hume’s (1739 p. 252) notorious inability to observe himself and Wittgenstein’s (1922 5.632) subject being no part of the world. But above all, it explains why even these beliefs pose no problems of self-reference — nor, given insight, of self-knowledge. And why, to have them, we still need no concepts of the self or the present: i.e. no beliefs, conscious or otherwise, about what distinguishes selves and times from each other and from other things, or oneself and the present from other selves and times. 8 How ‘I’ and ‘now’ refer So far so good. It now only remains to show that even when put into words, our subjective beliefs still pose no problems of reference, and call for no concepts of the self or the present. What does it take to put my subjective beliefs into words? First, of course, I must want to do it, and must have beliefs about the right words to use. But what beliefs? What beliefs about ‘I’ and ‘now’ will make me use those words to express a belief like B(IN)? The right sentence to express a belief is one that shares its truth conditions and hence its referents. But every subjective belief refers to whomever has it and when — and causal contiguity will make that belief make only that person express it, and only at that time. So the right sentence to express it is one that refers to whomever produces it and when. And, | believe, the words which make sentences like S(IN) do that are ‘I’ and ‘now’. That's the belief about those words which makes me use them to express my subjective beliefs. And all I need to do so is a habit, and some more subjective beliefs. The habit is simply this: my desire to express a belief habitually causes me to use ‘I’ when the belief I want to express is first person, and ‘now’ when it’s present tense. Given causal contiguity, that habit will automatically make any token ‘I’ and ‘now’ which it makes me produce, at any time Y, refer respectively to Hugh Mellor and to ¥. The habit needn’t of course be conscious. Native speakers needn’t consciously choose the words they use. But even when I do choose ‘I’ and ‘now’ consciously, 1 don’t choose them for their alleged reference-fixing ‘senses’: for, as we saw in 3, I would choose them even if I lacked all sense of the time and of who I was. I choose them simply because I believe them to be the right words to use for the subjective beliefs I wish to express.Tandnow 29 And what makes this belief true is just that everyone shares my habit of using those words to do this. That’s what makes all our ‘I’ and ‘now’ tokens refer, in sentences like S(/N), to those who produce them and when. And we understand each other because we believe truly that that’s what they do. That true subjective belief is what makes us rightly take each other’s ‘I’ and ‘now’ tokens to refer to those who produce them and when: and that’s what makes them the right words for me to use for that purpose (Lewis 1969). A knowingly shared habit is therefore all it takes for me to use ‘I’ and ‘now’ correctly and successfully at any time Y. I still don’t need a concept of the self or of the present; nor need I believe that my ‘now’ refers to a present now, or to Y, or that my ‘I’ refers to my /, or to Hugh Mellor. To refer as they do, those tokens need no such problematic senses. All they need is our habit of producing them when we want to express our subjective beliefs: beliefs whose linguistic expression therefore poses no more metaphysical, semantic or epistemic problems than they do themselves — namely, none at all.3 Consciousness and degrees of belief 1 Many of our beliefs come by degrees. Do beliefs about beliefs, especially beliefs about our own present beliefs, do so? In other words, do we have ‘second order’ degrees of belief? Skyrms (1980b) defends the idea that we do ‘against charges of inconsistency, illegitimacy and triviality’ and goes on to show its ‘theoretical usefulness in connection with the laws of motion for rational belief’ (p. 109). I go further. The idea of second order beliefs is not only legitimate and useful, we positively need it in order to provide a theory of conscious belief. And with this theory we can ward off a recent attack on the theory that degrees of belief are subjective probabilities. Combining these two theories seems, however, to have some unattractive consequences; but none, I believe, that need force us to reject them. In what follows I develop and defend these three claims in turn. In Part I, I develop a theory of conscious belief, and extend it to take in degrees of belief. In Part II, I use this theory as part of a general defence of subjective proba- bility as a measure of degrees of belief. (I anticipate this result from time to time in Part I by taking degrees of belief to be subjective probabilities. But that is only to simplify the discussion; it is not essential, and does not beg the question.) In Part III, I show how the theories of Parts I and II together seem to entail that we consciously believe all the consequences of our conscious beliefs. This clearly false conclusion in fact follows, not from these theories, but from the easy but erroneous idea that we always believe definite proposi- tions, i.e. that the contents of all our beliefs determine definite truth conditions for them. Not, however, wishing to pretend that I have worked out any better idea, I shall continue to use the term ‘proposition’ for the content of a belief. 2 Ramsey (1926) showed how to measure the degree of a belief by measuring how strongly its owner is disposed to act on it. Ramsey apropos of its degree, and Braithwaite (1933) in general, identify belief with this disposition to act rather than with any conscious feeling of conviction. Feelings of conviction 30Consciousness and degrees of belief 31 may accompany belief, but they need not: ‘the beliefs we hold most strongly are often accompanied by practically no feeling at all’ (Ramsey 1926 p. 65). For example, ‘I believe quite thoroughly that the sun will shine tomorrow, but experience no particular feeling attached to the proposition believed’ (Braithwaite 1933 p. 142). Feelings may no doubt be among the causes and effects of gaining or losing a belief, but it is the disposition which directly determines how the believer will behave. The disposition is what provides the proximate explanation of his actions. This might be denied, on the grounds that, as Moliére thought, it is vacuous to explain events by saying that things — or people — are disposed to make them happen. But this is not always so. The most reputable explanatory properties of people and things are just conjunctions of dispositions to bring about the events they explain (see chapter 6). Even if a person’s belief were just a disposition, using it to explain his behaviour would be no worse than using his weight to explain his effect on the bathroom scales. If anything, it should seem better to those who suspect dispositional explanation of vacuity, since a particular belief is not a disposition to any specific kind of activity: what its owner does will depend also on his desires and his other beliefs. So the beliefs and desires which are used to explain an action cannot be defined by it in the simple way in which Moliére complains that ‘dormitive virtue’ can be defined. This is indeed an important fact about beliefs, and one I shall have need of later on. Meanwhile it will be convenient to mark the fact immediately by calling beliefs ‘quasi-dispositions’, and Ramsey’s and Braithwaite’s theory of them an ‘action’ theory rather than a dispositional theory. It is in fact neither vacuous nor easy to characterise beliefs by the actions which, along with desires, they generate. In particular, it is not easy to say in general how changes in belief affect behaviour. Fortunately, we shall be con- cemed only with variations in the strength of belief, rather than in its content; and their effects are relatively easy to state. What they affect is betting behaviour, broadly conceived. As Ramsey put it, ‘all our lives we are in a sense betting. Whenever we go to the station we are betting that a train will really run, and if we had not a sufficient degree of belief in this we should decline the bet and stay at home’ (Ramsey 1926 p. 79). And as the degree of our belief in the train’s running gets less, we may add, so the journey has to matter more to us to get us to the station. This is not meant to imply that we consciously calculate odds whenever our degrees of belief affect our activity. Degrees of belief can influence quite un- selfconscious actions. That is indeed how degrees of belief are best revealed,32 Consciousness and degrees of belief since ex hypothesi such actions will not be being done deliberately in order to mislead. It is precisely because an action theory does not invoke conscious feelings of conviction that it can readily allow beliefs to explain actions done while the agent’s conscious thought is miles away. And so it should: that is how most actions are done. Crossing a two-way street in Britain, for example, I nearly always look right first, that being the direction | most strongly believe traffic on my side of the road will be coming from; but this thought crosses my con- scious mind much less often than I cross British streets. Animals likewise, we suppose, can act on beliefs without having to be conscious of them. ‘Often enough, my cat’s behaviour makes it clear to me that he believes he is about to be fed’, says Jeffrey (1983 p. 70), and I believe him. At least, I believe him more strongly than I believe his cat has any conscious thought about the matter. What is more, the animal could just as easily make it clear to me that he is none too sure of being fed, less sure for instance than he is of the prospects to be opened up by stalking a nearby thrush. In short, we readily grant some animals degrees of belief; and we grant them these degrees more readily than we grant them conscious convictions. It is a virtue therefore in Ramsey’s theory to let them have the one without the other. 3 Virtues, however, have their price. The price here is that an action theory does not as it stands make sense of conscious belief. It is not, as Ramsey (1926 p. 65) claims, ‘observably false ... to suppose that the degree of a belief is something perceptible by its owner’. On the contrary, their being perceptible to us is what makes immediately obvious the truth of Ramsey’s other claim that ‘the beliefs we hold most strongly are often accompanied by practically no feeling at all’. If belief were a feeling of conviction, the phenomenon of conscious belief would pose no problem, since kinds of consciousness are just what kinds of feelings are. But dispositions to act are not kinds of consciousness, and nor are the quasi-dispositional states of belief and desire which, according to the action theory, cause them. Such states of mind as these will not automatically intimate their presence to their owners as feelings do. Nor can we credibly claim a general ability to become conscious of our own properties, whether they be physical or mental. Fortunately for the medical industry, for example, I can- not detect by introspection either what my blood group is or whether I am colour blind. My beliefs, on the other hand, I can call into my consciousness almost whenever I like (doubtless with Freudian exceptions). The Humean idea of belief as a kind of self-intimating feeling still appeals precisely because itConsciousness and degrees of belief 33 accounts immediately for this fact. And until the action theory can produce an alternative account of what conscious belief is and how it occurs, that Humean idea, for all its defects, will never be properly scotched. Where should an alternative to the Humean idea of conscious belief be looked for? Armstrong, who espouses an action theory of belief, maintains that ‘an account of having a belief before the mind, as a current content of consciousness, does not demand development of the theory of belief but rather of the quite general notion of consciousness’ (Armstrong 1973 p. 22). I disagree: the contents of consciousness are too diverse. The action theory is after all forced on us in the first place because even conscious belief is so obviously unlike a feeling. It seems evident to me that conscious belief should be explained by relating it to its ‘unconscious’ action-producing counterpart, not by lumping it in a general theory of consciousness with items as unlike it as pains, visual sensations and emotions are. That at least is how I shall set about explaining it; it will be time to consider Armstrong’s preferred alternative if and when it ever appears. 4 need however to say a little more about conscious belief before offering a theory of it. I take it that a man’s mind changes in some way when he acquires a conscious belief, even if he acquires it only by becoming conscious of a belief he already has. Suppose I walk towards the back of a room in order to get out of it. I walk that way because that is where I most strongly believe the exit to be. But this need not be a conscious degree of belief: consciously, I may be completely preoccupied with my reason for leaving the room. If I do start to think consciously about where the exit is, I shall come thereby into a new state of mind, different from the one which would in any case propel me towards the door. For this new state, of conscious belief in the exit’s whereabouts, I need a new name, in order to discourage the idea that it is just the quasi-dispositional belief state plus consciousness. This conscious state is actually what Hume (1739 p. 629) called simply ‘belief, which is a term that every one sufficiently understands in common life’; and no doubt the conscious state is still what most people think of as belief. But since our account of belief is going to start from the basic concept of the action theory, ‘belief’ is best reserved for that ‘unconscious’ quasi-disposition. As in my earlier paper on conscious belief (Mellor 1978), therefore, I follow Price (1969 p. 189) and others in using Hume’s other term, ‘assent’, to refer to conscious belief. The term is not ideal, as I remarked on first adopting it: in particular, its connotations of a public display of acquiescence need to be discarded. Still, as Hume said, ‘provided we34 Consciousness and degrees of belief agree about the thing, ’tis needless to dispute about the terms’; and the mental state I have in mind is as familiar in common life today as it was to Hume, despite having proved so elusive to analysis. Throughout what follows, therefore, it must be kept in mind that this familiar inner state is what I mean by ‘assent’, not any of its outward manifestations in behaviour; and by ‘dissent’ I shall likewise mean the inner state of conscious disbelief. Assent, thus understood, evidently comes by degrees as much as plain belief does. Assenting to something is at least consciously believing it more strongly than its negation, and is quite often not much more than that. While between assent and dissent, as between belief and disbelief, lie the intermediate degrees of doubt, in this case of conscious doubt. In order to talk about all these conscious states together I shall use ‘assent’ as ‘belief’ is also used: namely, not only for one extreme of this family of mental states, but also as a name for the family itself, ranging in degree from what we may call full assent, through increasingly sceptical shades of doubt, to outright dissent. Not only can we assent in varying degrees to propositions of all sorts, we incessantly do. Almost whenever we are conscious, that consciousness includes assenting to something, or dissenting from it or consciously doubting it. This aspect of our consciousness moreover is what gives us an almost instant access to nearly all our own beliefs, which we have to no one else’s. | may see some of your beliefs reflected in your behaviour (as Jeffrey sees his cat’s), but I need not watch my own actions to get my own beliefs and doubts into my consciousness. Suppose for example I hear you put to me almost any question, ‘nh?’ which I understand. I will at once be made conscious, as you will not, of whether I believe A, disbelieve it or doubt it: i.e. | will at once assent to A in some greater or lesser degree. Now I may of course be wrong, not only about h itself but about the strength of my own belief in A. That is, not only may I assent to h when h is false, | may assent to A when I do not believe it. The degree of my assent, in other words, may differ markedly from the degree of plain belief which my unselfconscious actions would reveal to others. And it is an important fact that when this happens, we take it to be a case of self- deception: what we conclude is that I do not really believe what I consciously believe. This shows both how assent is thought to involve a fallible perception of belief, and how right the action theory is to take the unconscious state, and not assent, to be the paradigm of really believing something. Self-deception about belief is, I suppose, almost always possible. But it is not common; and, right or wrong, my assent does at least tell me something about my belief, which it does not tell me about anyone else’s. This is the familiar fact, referred to in 3, of our privileged access to our own beliefs.Consciousness and degrees of belief 35 This is the fact that an action theory needs to account for by construing assent also as a disposition or quasi-disposition to action. 5 The cue for an action theory of assent is given by the fact that some behaviour needs assent, not just plain belief. In particular, linguistic behaviour needs assent. To talk or write to someone I have to be conscious of what I am talking or writing about. My plain belief, that traffic here keeps left in two- way streets, may steer me safely through a town without it becoming a conscious belief, but it will not steer me through a conversation on the subject. For that I need assent. I do not mean that I need to assent fully to everything I say, since I can of course choose to lie. But what I lack when I lie is not consciousness of my state of belief. On the contrary, I have to be as conscious of what I believe in order to lie about it, as I do in order to tell what I believe to be the truth. What I lack when I lie is the degree of assent which is implicit in the remarks I choose to make. To lie, I must dissent from what I assert, or at least consciously doubt it more than my speech conveys. Plain disbelief is not enough. Some degree of assent therefore is needed for every assertion. But assent is not a simple disposition to assert, as the case of lying shows. What I say depends not only on what I assent to, but on how I desire to affect my audience. I shall only assert what I assent to if I want to tell my audience the truth; otherwise I shall deny it, equivocate or say nothing at all. So no degree of assent, however high, can be defined as a disposition to make any specific assertion, any more than plain beliefs, however strong, can be defined as dispositions to do any other specific thing. Assent, therefore, like belief, at least includes a quasi-disposition to act — but only to act in special ways, such as conversing, which call for it. Now if. we actually identify assent with such a quasi-disposition, we can spare ourselves the unhopeful search for a special kind of consciousness to accompany, or partly constitute, assent. We already know that no feeling or sensation does so: the whole point of the quotations from Ramsey and Braithwaite in 2 was to invite us to make some strong belief conscious, i.e. to generate a high degree of assent, and then consciously to note the absence of any particular feeling. Our problem then will be to relate these two quasi-dispositions, assent and plain belief. The problem is especially acute for those of us who wish to follow Ramsey in measuring the strength of a belief by the believer’s choice of odds for a bet on its truth. For literally choosing odds, like talking, calls for consciousness of the proposition in question, i.e. it calls for assent, and not just36 Consciousness and degrees of belief for belief. We have to give some reason for assuming that a conscious choice between explicitly risky alternatives will reveal the strength of beliefs of which we need not, even in action, be conscious at all. Nor even perhaps, in the case of animals, be capable of bringing to consciousness. We have credited animals with beliefs, and with beliefs of variable strength, but not with what it takes to choose odds, not with degrees of assent. Why should the strength of an assent of which they are incapable be supposed to measure the strength of the essentially unconscious belief that they actually have? 6 I propose to solve these problems with the simplest possible action theory of assent: namely, that assenting to a proposition is believing one believes it. | put this theory forward first in Mellor (1978), where I deliberately confined it to the qualitative case of full belief and full assent. But since belief and assent do in fact come by degrees, the theory must be shown also to accommodate that fact; and this is my present task. First, however, | should summarise the quali- tative theory, correcting some errors in the original version and dealing with some objections that have been made to it. I write ‘Brah’ for a person or animal a believing a proposition h at a time 7. Where the identity of a, ¢ or h is immaterial, I abbreviate this as convenient to “Bth’, ‘Bah’, ‘Bh’, or just ‘B’. Disbelief I take to be belief in the negation, which I write ‘B~h’: ‘~Bh’ means merely that h is not believed, which might mean disbelief, or doubt or merely a lack of any attitude at all to A. Originally I took a believing at f that he believes h to be BtaBtah (Mellor 1978 pp. 90-1), but this is wrong. Believing one believes something is an essentially indexical state: it is a believing at f that he believes h now. This is not the same as his believing that a believes h at t, as Perry (1979) has shown, even though he is a and ¢ is the present time. For instance, a may have forgotten the time, or even who he is, and not believe Brah under any true non-indexical description of t and a. But he can still believe that he himself, whoever he is, believes h now, whatever the time now is; and this is the state that concerns me. Since this state of believing one believes is indexed in this way to its owner and the present time, it will suffice for my purposes to write it simply as ‘BtaB*h’, or ‘BB for short, the ‘*’ serving to indicate its indexicality in these two respects. Then if A is the state of assent, as B is of plain belief, my thesis is that (1) A= BB*.Consciousness and degrees of belief 37 7 (1) involves two claims: (1a), that A entails BB*; and (1b), that BB* entails A. My argument for (1a) may be summarised by the following excerpts from Mellor (1978), duly modified to correct the error noticed above. There is prima facie such a thing as believing one believes something. I have beliefs about all sorts of things: why not about my own beliefs? [Moreover] we recognise a difference [between BB*h and Bh), as our concept of a state of self-deception shows ... A husband, we suppose, can (subconsciously) believe his wife to be unfaithful, while (consciously) believing that he believes nothing of the sort. (p. 91) If we have such states [as BB*A) at all, they surely occur when. we assent to propositions ... Assent is the conscious belief in h that is required for the sincere affirmation of h’s truth. In coming to assent to A I have perceived (or in cases of self-deception, misperceived) my belief in h: if that does not involve believing one believes h, what does? (p. 92) In the original discussion, the following counter-example was proposed to this rhetorical question. Suppose I am reluctantly persuaded, e.g. by a psycho- analyst, that I subconsciously believe some nasty proposition about my parents. Since it makes no odds to the argument, I shall for ease of exposition represent this as subconsciously disbelieving its negation, the nice proposition h. Then, despite the analyst, I still assent to h; but, it was alleged, I no longer believe I believe it, which shows that (1a), and therefore (1), is false. Not so: the case has been misstated. What we really have here is my assenting to a proposition A and also assenting to the indexical proposition ~B*h (= I don’t now believe h). That is not Ah and ~BB*h as was alleged; it is Ah and A~B*h, which according to (1) is BB*h and BB*~B*h, two perfectly compatible states of mind. Now I had indeed conjectured (Mellor 1978 p. 92) ‘that we have in fact no more distinct states [BB*B*h, BB*B*B*h] etc.’ beyond BB*h; and this case perhaps shows that we must occasionally allow BB*B*h to be a state distinct from BB*h. But my chief contention, (1a), survives unscathed: assenting entails believing one believes. The converse contention, (1b), is less obvious. If Bh can exist without my being conscious of it, as on an action theory it clearly can, one might ask why BB*h cannot do so. And if it can, it can presumably exist without assent, in which case (1b) is false. But this suggestion misses the whole point of (1). The thesis is, after all, that for Bh to be conscious just is for the state BB*h to38 Consciousness and degrees of belief occur. For that state in turn to be conscious would therefore be for BB*B*h to occur. But with the possible exception just noticed, I see little reason to suppose that there is any such state distinct from BB*h itself. And of course this account does not invoke consciousness in saying what assent is. That is a virtue, not a defect, since its very object is to say in other terms what this elusive kind of consciousness is. What a defender of (1) does need to do is, positively, to use it to explain why the kinds of behaviour that need assent do so, i.e. why they need BB* and not just B; and, negatively, to explain away putative counter-examples. In my (1978) I undertook both of these tasks; here I briefly summarise the results. Positively, I used a couple of Gricean truisms about language to show that sincerely communicating any h calls for BB*h, not just Bh. (What I originally claimed in Mellor (1978 VII) was that communication needs BBh, but in fact the token-reflexive state BB*h is what the argument really shows to be needed.) In being told that A, the hearer is intended by the speaker to be convinced that the speaker himself believes h as he speaks. To have this intention the speaker needs, if he is sincere, to believe that he himself now believes h, i.e. he needs to be in the state BB*h. (If he is being insincere, the state he needs to be in is BB*~h.) And if (1) is true, that fact explains why he needs to be in the state Ah (A~h), because it is the very same state. The thesis does therefore explain why conversation needs assent, not just belief. To dispose of most proposed counter-examples to (1b), it suffices to insist on the essentially indexical, present-tense nature of assent. Tenseless beliefs about beliefs, beliefs of tenses other than the present tense, beliefs about the beliefs of other people: none of these is assent, and any of them could exist without assent, for all (1b) says. The suspicion that BB* could do so stems largely from confounding it with other sorts of belief about beliefs. But one apparent counter-example to (1b) is properly indexed: namely, self-deception itself. A man deceives himself if he believes he believes h when he does not believe h: i.e. BB*h&~Bh. But we should not normally say that he was self-deceived about his belief in h only when that belief was conscious. So it seems that he could be in the state BB*h without being in any degree of the state Ah; so (1b) is false. In fact this is not so. The state of self-deception that can exist even when the man is not assenting to h is not in fact BB*h&~Bh. Now in saying this I am not just trying to rescue (1b) by stipulation. Recall that our objective is to extend the action theory to cope with assent, and the basic concept of belief to be used must therefore be that of the action theory itself. We have already seen that it differs from everyman’s concept in not implying consciousness; itConsciousness and degrees of belief 39 differs also in not applying to this state of self-deception. This is because what the action theory calls belief is a quasi-disposition (see 2 above): i.e. how a belief displays itself in action depends on what else its owner believes and what he desires. But this is not true of self-deception and self-knowledge. They are straight dispositions. Like fragility and Moliére’s ‘dormitive virtue’, these states always display themselves in the same way, by which therefore they can be defined. Regardless of what else he believes and desires, anyone who assents to (dissents from) h only when he believes it is, by definition, a man who knows (deceives) himself in this respect. An action theory of belief that can cope with assent is thus automatically equipped to cope with self-knowledge and self-deceit. The man who believes h, and knows himself in this respect, is a man disposed to assent to h, whether he is actually assenting to it or not. This indeed is a state distinct both from just believing h and from actually assenting to it, even though it is definable in terms of those two states. So there are three states to be accounted for altogether, but with the addition of our thesis (1), the action theory can account for all of them: the basic quasi-disposition, Bh; assent, BB*h; and the disposition, to BB*h (if BB*hvB~B*h) only if Bh. This disposition is not a belief as action theory construes belief. A fortiori it is not the belief BB*h&~Bh; so the fact that it can exist without AA is no objection to the thesis (). 8 Although self-knowledge is definable in terms of belief and assent, it is nev- ertheless a distinct, and a real, property of those that have it. It is a property distinct from belief and assent in just the way the fragility of a glass is a property distinct from its being dropped and its breaking. And, like fragility, self-knowledge is a real property in the sense, for example, that changes in it are real events with causes and effects (see chapter 6.1). It has in fact a causal mechanism, whereby whenever one’s belief about A comes to consciousness, the state Bh will cause the state BB*h, and the state B~h will cause the state BB*~h, It is the means whereby, as Ramsey (1926 p. 67) put it, ‘what determines how we should act determines us also directly or indirectly to have a correct opinion as to how we should act’; only not, as Ramsey thought, ‘without its ever coming into consciousness’. The mechanism of self-knowledge is like the mechanism whereby, when something we are looking at is red, it causes us to believe it’s red, and when it isn’t, to believe it isn’t. In short, self-knowledge is a species of perceptual ability, which enables facts to cause us to believe in them. That is, its causal mechanism is a sense, just as eyesight is a sense. Only, since what it informs us40 Consciousness and degrees of belief of is one of our own inner states of mind, it is an inner sense. It will be embodied entirely within the brain, because that is where both its objects, B, and the perceptions it delivers, BB*, are embodied. Its organs are therefore naturally unfamiliar to us, because they are less visible than those of our outer senses; its very existence, indeed, is for this reason quite easily overlooked. This inner sense is nevertheless crucial to my account of the relation between belief and assent, and in particular of the relation between first and second order degrees of belief. I must therefore beg leave to expatiate on it to some extent. I mean the claim, that we have an inner sense, literally, not just metaphori- cally; and I defended it at some length in Mellor (1978 IX), where I took the liberty of calling it ‘insight’. Insight is the mechanism of the privileged access that we have to our own beliefs, and as such it completes in principle the account, with which I said in 3 the basic action theory must be supplemented, of the phenomenon of assent. Details of the working of our insight, like the working of our other senses, are for physiological psychology to supply; what matters here is that we recognise its existence, and thus that there is nothing either infallible or scientifically ineffable about introspecting our own beliefs. When we introspect beliefs we are simply perceiving them with a special sense, just as with other senses we perceive visible and audible aspects of the outside world. Now the outside world of course contains more visual detail than meets the naked eye. In the same way, our minds no doubt contain more cognitive detail than meets our unassisted insight. The extreme case, corresponding to literal blindness, would be that of animals who have beliefs but have no insight at all. Since they lack the means of perceiving their own beliefs, they cannot assent in any degree to what their actions show them to believe. This in tum explains why they have no language, being incapable of Gricean intentions to commu- nicate their beliefs, and it does so without depriving them at all of sensations, emotions and other important aspects of consciousness which languageless animals may well possess. People, however, are not internally blind. We do have insight; but it is not perfect, any more than our eyesight is. At any time, most of our beliefs are not being perceived by us; and those that are will not be perceived in every detail. Thus, to anticipate an observation that is going to cause trouble in Part III, we are never aware of all the consequences of our conscious beliefs, not even of all those consequences which we do in fact believe. I believe, for example, that to drive quickly from Cambridge to London I should start off along Trumpington Road; and given my other geographical beliefs, it followsConsciousness and degrees of belief 41 that I should turn left from the end of Lensfield Road. But I can easily be con- scious of the first of these beliefs without being conscious of the second. Moreover, as we have seen, insight, like eyesight, is fallible. People can and do deceive themselves. And once we recognise that insight is just another sense, there is no more mystery about recognising self-deception than there is about recognising colour-blindness. Naturally neither state is directly percep- tible to its owner. But both can be seen by others, and in essentially the same way, namely by comparing the patient’s faulty perception with the thing perceived. And (pace Davidson 1970b and others) there is no special mystery about perceiving people’s beliefs and assentings. Mostly we infer them from what we know they have seen and heard. For the rest, we infer people’s beliefs from their unselfconscious actions, and what they assent to from their unstudied speech. That is why, when someone’s seemingly sincere speech makes too little sense of the rest of his behaviour, we infer that he is self- deceived - as when we listen sceptically to the subconsciously suspicious husband’s eager rationalisations of his incessant phone calls home. Admitting the fallibility of insight is thus by no means admitting wholesale scepticism about its deliverances. Insight we know to be no less reliable than our outer senses: for one thing, they themselves rely on it in conscious obser- vation, since that gives us assent as well as plain belief. And we know that our senses are reliable, that people mostly see things as they are and believe what they assent to, because we can see that they do. The reliability of our senses is a straightforwardly observable fact about the world of which we, and they, are perceptible parts. It is important to realise this, and moreover to realise that, as Grandy (1980) remarks, there is no epistemologically vicious circle or regress involved in it. It is because our senses are reliable that our seeing that they are amounts to our knowing it: i.e. it is a true belief got by a reliable process (Ramsey 1929b p. 121). We need not prove a priori that our senses are largely free of error before using them to confirm the fact (see chapter 15.5). It follows in particular that we need no special or a priori justification for assuming that insight generally reveals to us the degree of our plain beliefs. When Ramsey assumed that a conscious choice of betting odds generally reveals the degree of our unconscious belief, he assumed nothing that calls for a priori defence. He was simply, and rightly, taking for granted the observable reliability of one of our senses. 9 If (1) is true, and belief comes by degrees, assent should come by degrees in two distinct ways. Suppose my plain or ‘first order’ belief in A is neither full42 Consciousness and degrees of belief belief nor full disbelief, but is of some intermediate degree of doubt. For that to be the conscious degree of my belief in A is, I take it, for me to believe fully that I have this degree of belief in h. That is the natural way to construe degrees of assent. Insight, like our other senses, usually delivers full belief about what we perceive by means of it. But, again like our other senses, it need not. Just as poor eyesight may leave me unsure of the colour of an object, so poor insight may leave me unsure of the strength of my first order beliefs. In that case, if my degrees of belief are subjective probabilities, I should have a subjective probability distribution over possible values of my first order degrees of belief in h. And this is a second way, according to the theory, in which assent should be capable of coming by degrees. Is this really so? How, it will be asked, do we distinguish second order subjective probabilities from first order ones? What is the difference between being sure of doubting something and doubting that one is sure of it? It is natural to suspect second order probabilities of collapsing onto first order ones; and if they do so collapse, so does my theory. So the suspicion needs to be dispelled. Let us therefore consider some prima facie cases of second order probabilities, Suppose for example I know a coin to be fairly tossed but fear it may be double-headed. That is, I know the chance of heads is either 0.5 or 1, but am not sure which. Suppose I think each value equally likely. Here we have second order probabilities: degrees of belief in alternative values of the objective chance. (Actually, we have degrees of assent, in the first and straightforward sense; but therefore also degrees of belief if I am not self- deceived — and as the extra element of consciousness is immaterial, I shall discuss the case in the simpler terms of plain belief.) Now as far as my expectation of heads goes, these second order probabilities do indeed collapse onto first order ones. Their net effect on me is that I should bet on heads at 3:1 on, i.e. behave as if I thought the chance 0.75. So how do these supposed second order probabilities show up in my behaviour? They show up, for one thing, in how I should react to seeing the results of successive tosses: in particular, in how my degree of belief in heads resulting from the next toss should be affected by them. It affects, as Skyrms says, ‘the laws of motion for rational belief’. Tails just once, for example, should instantly reduce the degree of my belief in heads to 0.5, by proving that the coin is not double-headed; whereas it should have no such immediate effect if I had merely thought the chance 0.75. So the second order probabilities do not in this case collapse completely onto first order ones. But in this case the first order probabilities areConsciousness and degrees of belief 43 objective. There is an objective difference between a coin’s being double- headed and its merely being biased. Doubt about the one possibility is bound in the end to be distinguishable from certain belief in the other by how one reacts to evidence as to which is actually the case. When both first and second order probabilities are subjective, however, it is not so obvious how they are to be distinguished. If I doubt a proposition h, why should I react differently to fresh evidence for h or against it just because my doubt is second order rather than first? How, for example, will the alleged difference show up between (i) believing to degree | that I believe h to degree 0.75 and (ii) believing to degree 0.5 that I believe h to degree 1, and to degree 0.5 that I believe it to degree 0.5? On the theory I am propounding, the difference will be a difference in consciousness and hence, in particular, in linguistic ability. In state (i), I am conscious of a precise degree of doubt in h, which I can if I wish sincerely report. I can predict exactly how I should react to options whose desirability depends on the truth of A. I would have no hesitation now in picking precise odds for a bet on A’s truth. State (ii) is a very different, and a very odd state. What is odd about it is the discontinuous two-peaked second order probability distribution over the first order degrees of belief. Our senses, inner and outer, do not usually deliver such bizarre distributions. My eyes, for example, would rarely if ever make me see the colour of something to be possibly either pure white or pitch black but certainly nothing in between. They normally give me a roughly bell- shaped probability density distribution over a continuous range of colours similar in hue, brightness and saturation. That is, as the colour becomes more similar the degree of my assurance, that it isn’t the colour I saw, diminishes more rapidly. And similarly with insight. My consciousness of my doubts usually consists in a roughly bell-shaped probability density distribution over my first order degrees of belief. That is, the closer two degrees of my first order belief become, the more rapidly does the degree of my assurance, as to which one I really have, diminish. State (ii) seems bizarre simply because it is so rare. It is, nevertheless, a perfectly intelligible state, and there is no difficulty in saying how it differs from state (i). In state (ii) I should be less aware of what I believe about h. I could sincerely say that I do not totally disbelieve h. But all I could say positively is that, in unselfconscious actions to whose outcome A mattered, I should either act as if I were certain of A or dither in a way equivalent to betting on h at evens; and that, if asked now to bet on which of these two I would then do, evens are again the odds I should propose.44 Consciousness and degrees of belief The oddity of (ii) does not therefore count against our theory of assent. On the contrary, that theory is what tells us what it would be like to be in state (ii), and what people in it would be apt to do and say; and this is how we know that state (ii) is so rare. We know we ourselves are almost never in states like that, and that other people are not apt to do and say such things. Our theory of assent, in short, makes plenty of behaviour and respectable introspection avail- able to distinguish state (ii) from state (i) — and to distinguish both from the unconscious and inarticulate 0.75 degree of merely first order belief in h, which my unselfconscious action might reveal, with no degree (not even zero) of second order belief in / at all. So second order subjective probabilities do not collapse onto first order ones. They are introspectibly distinct states of mind and show up in quite different kinds of behaviour. But there are of course causal connections between them. They are connected one way by insight: with good insight, a degree of first order belief will produce a very high degree of second order belief in its true value. There are also connections the other way. Conscious thought is, amongst other things, a means of affecting our habits of action, i.e. our desires and the degree of our first order beliefs. After a shock from an electric kettle, considerable conscious reflection on the safety of the rewiring may be needed to make me pick it up again without hesitation, i.e. to drive the degree of my first order belief in its safety back towards 1. But it can be done: calculation can affect our beliefs just as observation can. I may not, by taking conscious thought, be able to add cubits to my stature, but I can add degrees to my first order beliefs. In other words, we are not only disposed to assent to what we believe: we are also disposed, on the whole, to believe what we assent to. I 10 If the thesis (1) is true, we will have second order subjective probabilities if we have first order ones. Higher orders are more doubtful. I conjectured in 6 that we generally lack third and higher order beliefs: insight does not show us all our inner states. But even first order subjective probabilities are contro- versial. One may admit that beliefs vary roughly in strength and still resist the full panoply of subjective probability. First order beliefs do not obviously come in such a finely graded range of strengths, and the second order range is still less obvious. The whole apparatus may reasonably be suspected of fantasy, of the heroic but unrealistic model building that seems, among some Scandinavian philosophers, to have superseded the saga-making of old.Consciousness and degrees of belief 45 Kyburg (1978) recently marshalled a notable attack on subjective proba- bility, and one can readily imagine his reaction to iterating it as I have done. But his attack seems to me mostly misdirected, and the rest of it can be met with the aid of the present theory of assent. With it we can in particular remove from the concept of subjective probability a pernicious equivocation of which Kyburg rightly complains. Subjective probability is mostly presented as measuring how strong people’s beliefs actually are. But there is evidence, which Kyburg adduces, purporting to show that the strength of people’s beliefs does not in fact satisfy this measure. Faced with this evidence, subjectivists have been apt to claim only to be prescribing the measure on grounds of rationality, i.e. to say not that people do have probabilistic degrees of belief, but that they are irrational if they don’t. But subjectivists who say that cannot also claim to be doing psychology. In particular they cannot claim to explain away people’s actual agreement on objective probabilities as being merely the result of their condi- tionalising their subjective probabilities in response to shared evidence. Moreover, reference to rationality at this stage seriously confuses the question of what people’s degrees of belief actually are with the question of what they ought to be, given knowledge of chances or of other inductive evidence for or against their truth. So it is important on several counts to decide whether subjective probability is supposed to be a theory of real or merely of rational degrees of belief. The theory of assent will enable us to do just that. I must emphasise that in what follows I shall not be arguing for the full- blooded subjective view of probability. I do not believe for a moment that one man’s degree of belief is as good as another’s, nor that conditionalisation does in fact explain away the phenomenon of agreement on values of objective chance. It is true that my view fits Kyburg’s definition of ‘subjective probability’, since I do think ‘the assignment of a numerical probability ... [need] not reflect any known or hypothetical frequencies’. But Kyburg’s is an unreasonably broad definition: one can quite well be an objectivist without being a frequentist. On the other hand, I do wish to characterise objective probability by the constraints it prescribes for degrees of belief, so I have some interest in defending subjective probability: namely, that the statement of my views will be somewhat simplified if degrees of belief have a probability measure. 11 Kyburg does not deny that we think of belief as coming in rough and com- parative degrees. He would not dispute the examples I have given, and I need not therefore multiply or defend them. But it is still worth emphasising how46 Consciousness and degrees of belief deeply the concept is rooted in our everyday judgments. Subjective probability is not a radical innovation of theoretical psychology such as Freudian theories, for example, have proposed. It is no more than a quantitative development of a perfectly familiar state of mind which we freely and uncontentiously attribute to ourselves and other people all the time. The only question is why one should jib at such a development. Let us compare it, as Ramsey (1926 p. 69) did, with Newtonian mechanics. That is a quantitative development of rough and comparative ideas of force and mass, i.e. of what sets things moving and of how hard things are to move. No one jibs at this development and, though I admit it is often easier to measure forces than beliefs, the case for each is much the same. In each case we have a theory that postulates a continuum of degrees of a state whose rough gradation is a commonplace. Each postulation is supported by quantitative applications which link the quantity proposed with its causes and effects. Perception produces beliefs of various strengths which in tum combine with desires ditto to produce definite action even in conditions of uncertainty. Similarly, gravity and other causes produce forces of various strengths, which in turn combine with masses ditto to produce definite accelerations. Force and mass are quantitative quasi-dispositions to yield the behaviour they explain, just as belief and desire are. And the range of behaviour explained by the theory that invokes each of these pairs of states is what gives us reason to believe in their existence. The reason may look more impressive in the case of force and mass, but the difference is only one of degree, and the degree should not be exaggerated. One could easily level at Newtonian mechanics most of the criticisms brought against subjective probability. One could object especially to the factual underdetermination of Newton's theory. Its apparatus can easily be made to appear grossly disproportionate to the data which it is based on and which it is used to explain. A thing’s mass, for instance, is an infinite conjunction of the dispositions it has to accelerate under forces of different strength. Only one of these dispositions is ever displayed at any one time, and an infinite number will never be displayed at all. One might well ask what the point can be of attributing to a thing so extravagant a profusion of dispositions, and how one could possibly claim to know the truth of such attributions. The second, epistemological, question I shall not try to answer. The answer to the first question can only be that these Newtonian dispositions are supposed to be present, whether they are displayed or not, in order to be available to explain any of an infinite number of possible interactions with other things. That is, we suppose it to be true of the thing now that it would accelerate atConsciousness and degrees of belief 47 definite rates under all these definite forces, whether or not it will ever be exposed to them. Well, it may in the same way be true of me now that I would behave in definite ways in an infinite range of definite circumstances, whether I am ever in them or not. I may have a quite precise degree of belief, even though it will only show up in actions which, given my desires and other beliefs, a slightly weaker (or slightly stronger) belief would make me forgo. The degrees of most of my beliefs never will show up in action. But that will be no reason to deny their existence unless I am also to be denied Newtonian mass. Now I am not in fact claiming precise degrees for my beliefs, any more than I would claim to have a precise mass. Nor does the theory of subjective probability entail so rash a claim. Its mathematics indeed makes sense of indef- initely precise degrees of belief, just as Newtonian mechanics does of precise degrees of force and mass. That is because it is not for the mathematics to limit @ priori the precision of these states. Their precision of course is limited, as that of all quantitative states is (see Mellor 1967), but only a posteriori, by the several natures of the kinds of things involved. It is nonsense, for example, to give the mass of Everest to the nearest gram; but that does not prevent Newtonian mechanics applying to the Himalayas. Similarly, subjective probability is not seriously impugned by the obvious fact that most of our beliefs have imprecise degrees, and are more properly represented by intervals than by precise values of subjective probability (cf. Levi 1974). Imprecision is not a feature peculiar to degrees of belief, and its signifi- cance should not be exaggerated. Nor should its extent. We may quite easily have more precise degrees of belief than we think we have. Assent, we have seen, is a product of insight, and insight is no more precise a sense than eye- sight is. My unselfconscious choices might well display more definiteness in my first order doubts that I can in advance be consciously aware of. I may be conscious only of thinking something probable, i.e. that my degree of belief in it lies in the interval (0.5, 1), when my first order degree of belief actually only spans the narrower interval (0.7, 0.8). No one may know this, just as no one knows the mass of most things; yet, it may be so. 12 All this, however, is still only a preface to the main dispute. Kyburg’s chief complaint is not that subjective probabilities are too inactive, too numerous or too precise for the behavioural data they purport to explain. His complaint is that the Dutch book argument he considers fails to show that degrees of belief are subjective probabilities at all; and that, on the contrary, real betting behaviour shows them not to be. And to answer this complaint, the48 Consciousness and degrees of belief relation of betting to belief needs to be examined more closely than it usually is. We do not normally learn the strength of people’s beliefs by making them choose odds for bets, any more than we normally measure their masses by their accelerations under unit forces. We infer beliefs and masses alike from behaviour which is also affected by many other factors. Such inferences of course need knowledge of what the other factors are, and of how they interact with belief and mass respectively to produce the behaviour we observe. Moreover, we need this knowledge even to discuss how in principle these quantities could be directly and unequivocally measured: because, to measure them, we have to try and specify a situation in which the influence of the other factors we know of can be either eliminated or allowed for. Then we can take behaviour in that situation to be an explicit measure of belief, mass, or whatever else concerns us. That our specification will succeed, that the measure will be of the very quantity we want, cannot be guaranteed a priori or independently of theories in which the quantity already figures, and which tell us what other factors interact with it. It is not a matter of an arbitrary specification stipulating what we are to mean by degree of belief or mass: we already know most of what we mean, and our specification must accommodate itself to what we know. It is not, for example, by stipulation that IQ tests measure intelligence. If they do, it is because we already know that intelligence is one of the factors which affect their results, and that the other mental factors which might do so have, as a matter of fact, had their effects eliminated or allowed for. What these other factors are, and how their effects may be eliminated, is for our psychology to say; the information is not given us either by pure reason or by raw uninterpreted experience. So it is with the betting measure of belief. We know a belief’s strength affects its owner’s choice of odds, or quotient, for a bet on its truth, But we also know that this is not the only mental factor which affects the choice. So we must constrain the betting situation in order to eliminate or allow for the effects of these other factors. In Mellor (1971 ch. 2) I proposed constraints to that end: the bet was to be compulsory, with the opponent choosing its direc- tion and the stake after the quotient is fixed. The point of these constraints is to prevent the quotient being directly affected by a preference for particular stakes, by desire for a particular outcome, or by a like or dislike of the process of betting. The constraints moreover do this in a way that was explicitly designed to meet objections which Kyburg has now resurrected against the inept Dutch book argument that is unfortunately the only one he considers. That argument tries to use the least odds a man would accept for aConsciousness and degrees of belief 49 bet on h’s truth in order to measure his degree of belief in h — and this, as Kyburg reminds us, is nonsense if the man is to be made to bet, as he must be for a Dutch book to be made against him. ‘No odds can be unacceptable to a man who is compelled to bet in any case’ (Mellor 1971 p. 36). But it makes perfect sense for a man to say what odds he would choose if he had to bet on h with his opponent deciding subsequently which of them will win if h turns out to be true. That is a very plausible measure of his degree of belief in h, if he has one, whether he is willing to bet or not. It must be emphasised that the justification for placing these constraints on the betting situation has nothing whatever to do with making degrees of belief satisfy the probability calculus. The constraints are there simply to prevent the quotient being affected by factors other than the degree of belief it is supposed to measure. They may not, as a matter of fact, suffice to make the quotient measure nothing but degree of belief, but they are clearly at least necessary to that end. And they do suffice to set up a valid Dutch book argument. The only way a man who is compelled to bet, at stakes and in directions subsequently determined by his opponent, can prevent certain loss is by choosing ‘coherent’ quotients, i.e. quotients which are probabilities. That the loss would otherwise be certain follows from the fact that his opponent is also trying to win the bet: if he weren’t, it wouldn’t be a bet at all. But nor would it be a bet if one party were certain to lose; it would only be a pointlessly complex method of giving goods away. It is irrelevant to remark as Kyburg (1978 p. 162) does that professional gamblers make money quite rationally by offering non-probabilistic betting quotients. So they do, but they would not do so in situations constrained to reflect only their degrees of belief and not their greed. They would be coherent all right in the situation I have specified. The experiments Kyburg cites (p. 165), in which actual quotients close to 1 and 0 were not coherent, are equally inconclusive. His description of the experiments makes it clear that they did not satisfy the constraints needed for the quotients to measure only degrees of belief. The choice of quotients was clearly open to influence by some of the other factors I have mentioned, factors which would be especially likely to affect values close to 1 and 0, since these values greatly increase the possible gain and loss. Those experiments no more falsify subjective probability theory than a car’s needing an engine to keep going on the flat falsifies Newton’s first law of motion. 13 Still, it may be urged, even in the betting situation I have specified, a man could choose incoherent betting quotients. He will lose money, but he need not50 Consciousness and degrees of belief have had that intention; and if he did not, that intention cannot have been the cause of his behaviour. Lacking any other cause, then, we can only attribute it to his having non-probabilistic degrees of belief. We may of course suspect other unknown causes — but, after all, there may not be any. To insist a priori that there must be would simply beg the question in favour of subjective prob- ability. I remarked in 10 that many subjectivists admit that people may have non- probabilistic degrees of belief, and use the Dutch book argument only to show that such people are irrational. And that much the argument certainly shows: a man who has no desire to give away his goods should certainly not choose betting quotients which he knows will inevitably have that effect. But we can do better than this. With the aid of the theory of assent developed in Part I, we can show that degrees of belief really are probabilities, not just that they ought to be. First we should observe that the argument just given for the possibility of incoherent quotients only shows that degrees of belief might not be probabil- ities, not that they are not. It would only defeat an a priori argument for a probability measure, and although I hope my argument will prove compelling, I do not mean it to be a priori. The facts it appeals to, about assent and the psychology of betting, are entirely contingent. It could be defeated by an experimental proof that degrees of belief are not probabilities; but Kyburg, we saw in 12, has provided no such thing. Next, I have to say that the significance of betting for the measurement of belief has been uniformly misconceived in the literature, a misconception with which I have so far gone along. It is actually not at all like measuring masses by accelerating them, or even by weighing them. We cannot directly discover degrees of belief by observing quotients chosen for actual bets. The most we could discover directly in this way are degrees of assent, not degrees of first order belief. Choosing odds is a selfconscious activity, as I remarked in 5. What it reveals is what the chooser believes the degree of his first order beliefs to be. Usually he is right, for the reasons discussed in 8: insight is a generally reliable sense. But right or not, a gambler’s choice of odds does no more than report the result of his internal observation of the degree of his first order beliefs, an observation he can make without going on to gamble at all. This is why there is no point in making people bet in order to determine the degree of their beliefs. Thinking of odds or quotients is just a way of providing a uniform scale for reporting the deliverances of insight. It is like training people to report their feelings of warmth in degrees Celsius, or theirConsciousness and degrees of belief 51 visual sensations of colour in Angstrom units. Such a training has the virtue of sharpening our senses by providing an indefinitely precise and extendable vocabulary for stating our observations in. That such a vocabulary does have this virtue may be only a contingent fact, but it is a very familiar and impor- tant one. My ability to discriminate different degrees in one of my beliefs is undoubtedly improved by my habit of thinking what odds I should choose if I had no other reason for my choice. And not only different degrees of the same belief: the vocabulary of odds enables me to compare the strength of beliefs with widely different contents. Without this vocabulary, our perception of these inner states would certainly be much less precise. And because of the causal links between first and second order beliefs (see 9 above), the degrees of our first order beliefs might then well be less precise themselves. But what is the relevance of betting, if there is no point in actually doing it? How does the vocabulary of betting quotients acquire its virtue? Suppose I am thinking what quotients I should choose if I had no reason for my choice other than my degree of belief. I must still conceive myself to be choosing for a bet, albeit one constrained as specified in 12. If there were no bet, then even my degree of belief would give me no occasion for choosing one quotient rather than any other. And the occasion the bet supplies is simply one of possible but uncertain gain or loss. If either gain or loss were certain, I should again have no occasion to choose a particular quotient: my insight would be given no basis on which to answer my hypothetical question. So what I must suppose is that neither gain nor loss is certain in the circumstances; which means that the quotients must be supposed to be coherent, and therefore prob- abilities. I might of course sincerely produce incoherent betting quotients in reporting the degrees of my beliefs. But if I did, I should not be rightly reporting an irrational state of mind. Irrationality only lies in actually betting at such quotients in situations so constrained as to measure belief, and | am not actually doing that. But it is just because all such quotients would be equally irrational that they make no psychological sense as measures of my particular degree of belief. So the fact that I produce them shows, not that I am being irrational, but that, in introspecting my degrees of belief, | have made some error of measurement. It is as if I had insight also into my mass, and felt my- self to be more or less massive than my acceleration under Newtonian forces showed me to be. Newtonian mechanics tells me my introspection must have deceived me; and it will not do to retort that Newtonian mechanics might be wrong. So it is, but it was not that lightly overthrown; nor is subjective prob- ability overthrown by the exactly analogous experiments that Kyburg cites.52 Consciousness and degrees of belief Even if the inept design of these experiments had not made them inconclu- sive, therefore, even as introspections of degrees of belief, they would still have been completely insignificant. But what then would overthrow subjective probability, if not experiments of some such kind? I am not advocating subjective probability a priori, so something should be able to show if it is wrong. Well, consider what would show Newtonian mechanics to be wrong. Suppose certain bodies did not always accelerate as Newton requires under all forces. As a matter of fact, they don’t exactly, and not only for relativistic reasons. Different motions strip off different numbers of their surface molecules, and carry different amounts of air along with them. So the accelerations produced by different Newtonian forces are not, even at low speeds, related to each other by any one precise value of inertial mass. The utility, and the truth, of Newton’s theory consists in the accelerations being almost always related by values lying within some narrow interval (cf. Mellor 1967). If that interval became too wide for useful prediction, the concept of mass would eventually be discarded: Newtonian mechanics would have been falsified. But whilst we retain the concept at all, we retain the laws of motion by which its values, however imprecise, are determined. And so it is with degrees of belief. Our unselfconscious behaviour no doubt exhibits some quantitative inconsistency. | dare say I should not exhibit precisely the same degree of first order belief in imminent rain (say) in all the diverse unselfconscious actions which some degree of that belief might serve to explain. The best that can truly be said of me is that the degrees of this first order belief which would be exhibited in most of these actions lie within a certain interval of subjective probability. The wider the interval, the less useful the concept, and it is an entirely contingent matter that it has any use at all. My unselfconscious behaviour could be so erratic as to falsify the claim that I had any definite degree whatever of some particular belief. This is likely to be true, for example, of beliefs expressible only in languages I do not understand, such as the mathematical languages of most microphysics. 1 know I have no determinate degree of belief on those matters, ie. that my unselfconscious actions are in no way affected by any such state of mind. Consequently, although I could indeed be forced to choose precise odds for a compulsory bet on the truth of some such microphysical proposition, I know it would manifest no insight into the degree of any first order belief I have. But on matters I understand, my insight does tell me to avoid quotients outside certain intervals. There is a significantly narrow interval of subjective probability values within which I believe, with high second order subjective probability, my first order degree of belief to be confined. On that basis I canConsciousness and degrees of belief 53 predict, and generally claim to know, what my behaviour would be under quite a wide range of circumstances. Where that is the case, we have the same justification for applying the theory by which degrees of belief may, however imprecisely, be measured as we have in the case of mass for employing Newton’s laws of motion. As a matter of fact, therefore, many of our beliefs have usefully precise degrees. Of those that do, the Dutch book argument shows there to be a probabilistic measure. It is not a matter of rationality, but a matter of a theoretically based scale of measurement. Rationality comes in only later, in considering what degrees people should have of various beliefs in various circumstances. (And even that is mostly a matter not of rationality, but of chance.) Ii 14 Parts I and II have given reasons for thinking that assent is second order belief, that first and second order beliefs come more or less precisely by degrees, and that these degrees are subjective probabilities. I have not proved any of these propositions: I only incline to believe them because they enable us to explain sundry psychological phenomena. But that of course is not enough. I need also to check that their other consequences are equally acceptable. Some of the other consequences are both familiar and attractive. One is the explanation they provide of how we can fail to believe consequences of our beliefs. The explanation runs briefly as follows. Clearly no one believes any contingent proposition A to degree 1, i.e. no one behaves as if he would risk unlimited loss should h prove false for a penny gain should h prove true. So the qualitative state of full belief can only call for some high degree of belief, short of 1. The degree will no doubt be determined by the context: roughly, the least degree such that, in the context, no higher degree would affect anything the believer would be at all likely to do. The details are both tricky and important, but all that matters here is that the degree of a full belief can be less than 1. For then I can believe two things without believing their conjunction. Let p(k) be the probabilistic degree of my belief in any proposition k, and suppose there are two particular propositions A and i such that (2) pi) = pl) = 0.95. Suppose | also believe h and i to be independent, i.e. 3) p(h&i) = p(h)p(i) < 0.91.54 Consciousness and degrees of belief Suppose further that for some range of propositions k including h and i, (4) Bk EF pn, where 0.91i is necessary, so I believe it to degree 1. It follows from the probability calculus that I believe i at least as strongly as I believe A, i.e. (7) pth) $ p.! IFor all Jropositons handi, re 2 ba ph8-’,56 Consciousness and degrees of belief Now suppose p, is the subjective probability 1 need in the circumstances for full belief. Then, whatever p, is, (8) (p(h)>p,) - @O>p). That is, if I believe h, I must believe é: (9) Bhi-Bi. In other words, I must believe every consequence of each proposition I believe. This again, although not an attractive result, could be defended in the same way as its predecessor. It is not a matter of assent, only of first order belief construed as a quasi-disposition to unselfconscious action. My actions could of course show that I have changed my mind from believing / at one time to not believing i at another, or display the sort of inconsistency which makes my degrees of belief in A and i imprecise. Granting that, one could argue that no single action could reveal both that I definitely believe something and at the same time that I definitely fail to believe its logical consequences. So (9) might perhaps be accepted, albeit reluctantly and with foreboding. The foreboding is well justified. Consider our second order beliefs in h and i. ‘Bh’ and ‘Bi’ in (9) are of course abbreviations, as explained in 6, the refer- ence to the common time and believer being left out. The abbreviation is all right because (9) is a necessary truth: it holds of all believers at all times. So it holds in particular of myself at the present time, whatever time that is and whoever I am. That is, it holds of the indexical states B*h and B*i: (10) B*h |- BY. That is, B*h>B*i is necessary, so | believe it to degree 1. It follows that I believe B*i at least as strongly as I believe B*A, i.e. (11) p(B*h) < p(B*i). Now suppose p, is the subjective probability I need in the circumstances for full second order beliefs. Then, whatever p, is, (12) (p(B*h)>p,) |- (p(B*i)>p,)- and hi |- ~(h&~). So p(h>i)=1 | plh&~i)=0. So if h +i, plh) = p(h&i) < pli).Consciousness and degrees of belief 57 That is, if at any time I believe I now believe h, I must at the same time believe I now believe i: (13) BB*h |- BB*i. But that, given (1), entails (14) Ah Ai. In other words, if I assent to A, I must assent to i. We are being required to assent simultaneously to every logical consequence of any single proposition we assent to. This result cannot possibly be accepted. It is a result about conscious belief, not just about unconscious quasi-dispositions to action. No amount of theory could defend it against our consciousness of almost never consciously believ- ing all the consequences of one of our conscious beliefs. So unless there is a flaw in the argument leading to (14), one or other of its premises will have to go. Either assent is not believing one believes, or belief does not come by degrees of subjective probability. Let us survey the options. 16 There are two reasons for trying to keep the theory (1), that assent is believing one believes. One is that without it we shall have no theory of assent; nor, I suggest, of second order belief either. If believing one believes is not assenting, I cannot imagine what it is. But whatever it is, (13) seems to me barely more believable than (14). Believing one believes a proposition does not seem to entail believing one believes all its consequences, since something might be a consequence without one believing one believed it was. And that is the other reason for keeping (1): giving it up generates other problems without solving this one. The real trouble with (14) is (13), not (1). Perhaps the trouble with (13) is (12), i.e. lies in the theory of subjective probability. If so, it is not that our argument has credited beliefs with implau- sibly precise degrees. It nowhere depends on two contingent beliefs being of precisely the same strength. So long as my beliefs in A, i and in my first order beliefs in h and i are strong enough to be full beliefs, (12) follows. And imprecision in belief will not generally stop it being full belief. It might for propositions like those referred to in 13, expressible only in languages I barely understand. Perhaps there are some propositions of physics I under- stand so little I can neither fully believe nor fully disbelieve them. But (14) is no better when h and i are propositions so plain I must be able to believe them: e.g. h, that John was born in 1948; i, that he was born in a leap year. I can easily assent to A without thinking about / at all.